├── .gitignore ├── .project ├── .pydevproject ├── LICENSE ├── README.md ├── benchmark ├── GTX980_CUDA │ ├── sk_benchmark │ │ ├── process_forward_layers_timings.csv │ │ ├── process_memory_usage.csv │ │ ├── process_total_timings.csv │ │ ├── train_backward_layers_timings.csv │ │ ├── train_forward_layers_timings.csv │ │ ├── train_memory_usage.csv │ │ └── train_total_timings.csv │ ├── u_benchmark │ │ ├── process_forward_layers_timings.csv │ │ ├── process_memory_usage.csv │ │ └── process_total_timings.csv │ └── usk_benchmark │ │ ├── process_forward_layers_timings.csv │ │ ├── process_memory_usage.csv │ │ └── process_total_timings.csv ├── GTX980_OPENCL │ ├── sk_benchmark │ │ ├── process_forward_layers_timings.csv │ │ ├── process_memory_usage.csv │ │ ├── process_total_timings.csv │ │ ├── train_backward_layers_timings.csv │ │ ├── train_forward_layers_timings.csv │ │ ├── train_memory_usage.csv │ │ └── train_total_timings.csv │ ├── u_benchmark │ │ ├── process_forward_layers_timings.csv │ │ ├── process_memory_usage.csv │ │ └── process_total_timings.csv │ └── usk_benchmark │ │ ├── process_forward_layers_timings.csv │ │ ├── process_memory_usage.csv │ │ └── process_total_timings.csv ├── Old Benchmarks │ ├── E5-2697v3_OPENCL │ │ ├── sk_benchmark │ │ │ ├── process_forward_layers_timings.csv │ │ │ ├── process_memory_usage.csv │ │ │ ├── process_total_timings.csv │ │ │ ├── train_backward_layers_timings.csv │ │ │ ├── train_forward_layers_timings.csv │ │ │ ├── train_memory_usage.csv │ │ │ └── train_total_timings.csv │ │ ├── u_benchmark │ │ │ ├── process_forward_layers_timings.csv │ │ │ ├── process_memory_usage.csv │ │ │ ├── process_total_timings.csv │ │ │ ├── train_backward_layers_timings.csv │ │ │ ├── train_forward_layers_timings.csv │ │ │ ├── train_memory_usage.csv │ │ │ └── train_total_timings.csv │ │ └── usk_benchmark │ │ │ ├── process_forward_layers_timings.csv │ │ │ ├── process_memory_usage.csv │ │ │ ├── process_total_timings.csv │ │ │ ├── train_backward_layers_timings.csv │ │ │ ├── train_forward_layers_timings.csv │ │ │ ├── train_memory_usage.csv │ │ │ └── train_total_timings.csv │ ├── GTX980_CUDA │ │ ├── sk_benchmark │ │ │ ├── process_forward_layers_timings.csv │ │ │ ├── process_memory_usage.csv │ │ │ ├── process_total_timings.csv │ │ │ ├── train_backward_layers_timings.csv │ │ │ ├── train_forward_layers_timings.csv │ │ │ ├── train_memory_usage.csv │ │ │ └── train_total_timings.csv │ │ ├── u_benchmark │ │ │ ├── process_forward_layers_timings.csv │ │ │ ├── process_memory_usage.csv │ │ │ └── process_total_timings.csv │ │ └── usk_benchmark │ │ │ ├── process_forward_layers_timings.csv │ │ │ ├── process_memory_usage.csv │ │ │ └── process_total_timings.csv │ ├── GTX980_OPENCL │ │ ├── sk_benchmark │ │ │ ├── process_forward_layers_timings.csv │ │ │ ├── process_memory_usage.csv │ │ │ └── process_total_timings.csv │ │ ├── u_benchmark │ │ │ ├── process_forward_layers_timings.csv │ │ │ ├── process_memory_usage.csv │ │ │ └── process_total_timings.csv │ │ └── usk_benchmark │ │ │ ├── process_forward_layers_timings.csv │ │ │ ├── process_memory_usage.csv │ │ │ └── process_total_timings.csv │ ├── W9100_OPENCL │ │ ├── sk_benchmark │ │ │ ├── process_forward_layers_timings.csv │ │ │ ├── process_memory_usage.csv │ │ │ └── process_total_timings.csv │ │ ├── u_benchmark │ │ │ ├── process_forward_layers_timings.csv │ │ │ ├── process_memory_usage.csv │ │ │ ├── process_total_timings.csv │ │ │ ├── train_backward_layers_timings.csv │ │ │ ├── train_forward_layers_timings.csv │ │ │ ├── train_memory_usage.csv │ │ │ └── train_total_timings.csv │ │ └── usk_benchmark │ │ │ ├── process_forward_layers_timings.csv │ │ │ ├── process_memory_usage.csv │ │ │ ├── process_total_timings.csv │ │ │ ├── train_backward_layers_timings.csv │ │ │ ├── train_forward_layers_timings.csv │ │ │ ├── train_memory_usage.csv │ │ │ └── train_total_timings.csv │ └── i7-4790K_OPENCL │ │ ├── sk_benchmark │ │ ├── process_forward_layers_timings.csv │ │ ├── process_memory_usage.csv │ │ └── process_total_timings.csv │ │ ├── u_benchmark │ │ ├── process_forward_layers_timings.csv │ │ ├── process_memory_usage.csv │ │ ├── process_total_timings.csv │ │ ├── train_backward_layers_timings.csv │ │ ├── train_forward_layers_timings.csv │ │ ├── train_memory_usage.csv │ │ └── train_total_timings.csv │ │ └── usk_benchmark │ │ ├── process_forward_layers_timings.csv │ │ ├── process_memory_usage.csv │ │ ├── process_total_timings.csv │ │ ├── train_backward_layers_timings.csv │ │ ├── train_forward_layers_timings.csv │ │ ├── train_memory_usage.csv │ │ └── train_total_timings.csv ├── W9100_OPENCL │ ├── sk_benchmark │ │ ├── process_forward_layers_timings.csv │ │ ├── process_memory_usage.csv │ │ ├── process_total_timings.csv │ │ ├── train_backward_layers_timings.csv │ │ ├── train_forward_layers_timings.csv │ │ ├── train_memory_usage.csv │ │ └── train_total_timings.csv │ ├── u_benchmark │ │ ├── process_forward_layers_timings.csv │ │ ├── process_memory_usage.csv │ │ ├── process_total_timings.csv │ │ ├── train_backward_layers_timings.csv │ │ ├── train_forward_layers_timings.csv │ │ ├── train_memory_usage.csv │ │ └── train_total_timings.csv │ ├── usk_benchmark │ │ ├── process_forward_layers_timings.csv │ │ ├── process_memory_usage.csv │ │ ├── process_total_timings.csv │ │ ├── train_backward_layers_timings.csv │ │ ├── train_forward_layers_timings.csv │ │ ├── train_memory_usage.csv │ │ └── train_total_timings.csv │ └── usk_benchmark_512 │ │ ├── process_forward_layers_timings.csv │ │ ├── process_memory_usage.csv │ │ ├── process_total_timings.csv │ │ ├── train_backward_layers_timings.csv │ │ ├── train_forward_layers_timings.csv │ │ ├── train_memory_usage.csv │ │ └── train_total_timings.csv ├── benchmark_sk.sh ├── benchmark_u.sh ├── benchmark_u_small.sh ├── benchmark_usk.sh ├── clBLAS-BENCH │ ├── bench.sh │ ├── bench.txt │ └── bench.txt.old ├── i7-4790K_OPENCL │ ├── sk_benchmark │ │ ├── process_forward_layers_timings.csv │ │ ├── process_memory_usage.csv │ │ ├── process_total_timings.csv │ │ ├── train_backward_layers_timings.csv │ │ ├── train_forward_layers_timings.csv │ │ ├── train_memory_usage.csv │ │ └── train_total_timings.csv │ ├── u_benchmark │ │ ├── process_forward_layers_timings.csv │ │ ├── process_memory_usage.csv │ │ ├── process_total_timings.csv │ │ ├── train_backward_layers_timings.csv │ │ ├── train_forward_layers_timings.csv │ │ ├── train_memory_usage.csv │ │ └── train_total_timings.csv │ └── usk_benchmark │ │ ├── process_forward_layers_timings.csv │ │ ├── process_memory_usage.csv │ │ └── process_total_timings.csv ├── train_process_sk_2.prototxt ├── train_process_u_2.prototxt ├── train_process_u_2small.prototxt ├── train_process_usk_2.prototxt └── u_benchmark │ ├── process_forward_layers_timings.csv │ ├── process_memory_usage.csv │ ├── process_total_timings.csv │ ├── train_backward_layers_timings.csv │ ├── train_forward_layers_timings.csv │ ├── train_memory_usage.csv │ └── train_total_timings.csv ├── dataset_01 ├── .gitignore ├── process.sh ├── train.sh ├── train │ ├── labels │ │ ├── labels00000000.png │ │ ├── labels00000001.png │ │ ├── labels00000003.png │ │ ├── labels00000004.png │ │ ├── labels00000005.png │ │ ├── labels00000006.png │ │ ├── labels00000008.png │ │ ├── labels00000009.png │ │ ├── labels00000010.png │ │ ├── labels00000011.png │ │ ├── labels00000013.png │ │ ├── labels00000014.png │ │ ├── labels00000015.png │ │ ├── labels00000016.png │ │ ├── labels00000018.png │ │ └── labels00000019.png │ └── raw │ │ ├── 00.tif │ │ ├── 01.tif │ │ ├── 03.tif │ │ ├── 04.tif │ │ ├── 05.tif │ │ ├── 06.tif │ │ ├── 08.tif │ │ ├── 09.tif │ │ ├── 10.tif │ │ ├── 11.tif │ │ ├── 13.tif │ │ ├── 14.tif │ │ ├── 15.tif │ │ ├── 16.tif │ │ ├── 18.tif │ │ └── 19.tif ├── train2.sh ├── train_debug.sh ├── train_process_sk_2.prototxt ├── train_process_sk_9.prototxt ├── train_process_u_2.prototxt ├── train_process_u_9.prototxt ├── train_process_upsample.prototxt ├── train_process_usk_2.prototxt ├── train_process_usk_9.prototxt ├── validate_raw │ └── validate_raw.tif └── validate_target │ ├── labels00000002.png │ ├── labels00000002_cons.png │ ├── labels00000007.png │ ├── labels00000007_cons.png │ ├── labels00000012.png │ ├── labels00000012_cons.png │ ├── labels00000017.png │ ├── labels00000017_cons.png │ └── validate_target.tif ├── dataset_02 ├── .gitignore ├── process.sh ├── train.sh ├── train │ ├── labels │ │ └── train-labels00.tif │ └── raw │ │ └── train-volume00.tif ├── train2.sh ├── train_process_sk_2.prototxt ├── train_process_u_2.prototxt └── train_process_usk_2.prototxt ├── dataset_06 ├── .gitignore ├── config.py ├── config.pyc ├── fibsem_medulla_7col │ └── README.md ├── greentea_brew_tool.py ├── malis │ ├── .gitignore │ ├── README.md │ ├── __init__.py │ ├── __init__.pyc │ ├── make.sh │ ├── malis.pyx │ ├── malis_cpp.cpp │ ├── malis_cpp.h │ ├── setup.py │ ├── test_malis.py │ └── test_malis_nhood.ipynb ├── net │ ├── anisonet_euclid.prototxt │ ├── anisonet_malis.prototxt │ ├── anisonet_test.prototxt │ ├── net_test.prototxt │ └── solver.prototxt ├── netconf.py ├── netconf.pyc ├── network_generator.py ├── volume_slicer.py └── volume_slicer.pyc ├── dataset_08 ├── .gitignore ├── config.py ├── config.pyc ├── device_query.sh ├── net │ ├── net_test.prototxt │ ├── net_train.prototxt │ ├── net_train_malis.prototxt │ └── solver.prototxt ├── netconf.py ├── netconf.pyc ├── network_generator.py ├── process.sh ├── train.sh └── train_process.prototxt ├── evaluation └── fiji_script_isbi.bsh ├── malis_setup ├── draw_net.sh └── neuraltissue_net.prototxt ├── net_old ├── neuraltissue_process.prototxt ├── neuraltissue_solver.prototxt ├── neuraltissue_test.prototxt └── neuraltissue_train.prototxt ├── net_sk_2out ├── draw_net.sh ├── neuraltissue_net.prototxt ├── neuraltissue_net_malis.prototxt ├── neuraltissue_net_softmax.prototxt └── neuraltissue_solver.prototxt ├── net_sk_9out ├── neuraltissue_net.prototxt └── neuraltissue_solver.prototxt ├── net_u_2out ├── draw_net.sh ├── neuraltissue_net.prototxt ├── neuraltissue_net_malis.prototxt ├── neuraltissue_net_softmax.prototxt └── neuraltissue_solver.prototxt ├── net_u_9out ├── neuraltissue_net.prototxt └── neuraltissue_solver.prototxt ├── net_u_small ├── net_test.prototxt ├── net_train.prototxt ├── neuraltissue_net.prototxt ├── neuraltissue_solver.prototxt └── script.py ├── net_upsample └── neuraltissue_net.prototxt ├── net_usk_2out ├── draw_net.sh ├── neuraltissue_net.prototxt ├── neuraltissue_net_malis.prototxt ├── neuraltissue_net_softmax.prototxt └── neuraltissue_solver.prototxt ├── net_usk_3out ├── neuraltissue_net.prototxt └── neuraltissue_solver.prototxt ├── net_usk_9out ├── neuraltissue_net.prototxt └── neuraltissue_solver.prototxt ├── pygt_01 ├── loss.mat ├── net_test.prototxt ├── net_train.prototxt └── script.py ├── pygt_example ├── .gitignore ├── big_testnet.ipynb ├── draw_net.sh ├── netgen_example.ipynb ├── nopadding_example.ipynb └── reswnet.ipynb ├── pygt_fibsem ├── .gitignore ├── net_test.prototxt ├── net_train_euclid.prototxt ├── net_train_malis.prototxt ├── process.py └── script.py ├── pygt_isbi2012 ├── mknet.py ├── net_test.prototxt ├── net_train_euclid.prototxt ├── net_train_malis.prototxt ├── process.py └── train.py ├── pygt_snemi_aniso ├── mknet.py ├── net_test.prototxt ├── net_train_euclid.prototxt ├── net_train_malis.prototxt └── train.py └── pygt_uvisual ├── .gitignore ├── mknet.py ├── net_test.prototxt ├── net_train_euclid.prototxt ├── net_train_malis.prototxt ├── process.py ├── test_out.h5 ├── testnet.aux ├── testnet.bcf ├── testnet.log ├── testnet.run.xml ├── testnet.synctex.gz ├── testnet.tex ├── train.py └── trainnet.tex /.gitignore: -------------------------------------------------------------------------------- 1 | *.ipynb_checkpoints* 2 | *.caffemodel 3 | *.solverstate 4 | dataset_03 5 | dataset_04 6 | dataset_05 7 | dataset_07 8 | *.fuse* 9 | *.pdf 10 | *.ps 11 | -------------------------------------------------------------------------------- /.project: -------------------------------------------------------------------------------- 1 | 2 | 3 | project_data 4 | 5 | 6 | 7 | 8 | 9 | org.python.pydev.PyDevBuilder 10 | 11 | 12 | 13 | 14 | 15 | org.python.pydev.pythonNature 16 | 17 | 18 | -------------------------------------------------------------------------------- /.pydevproject: -------------------------------------------------------------------------------- 1 | 2 | 3 | Default 4 | python 2.7 5 | 6 | -------------------------------------------------------------------------------- /LICENSE: -------------------------------------------------------------------------------- 1 | Copyright (c) 2015, Fabian Tschopp 2 | All rights reserved. 3 | 4 | Redistribution and use in source and binary forms, with or without 5 | modification, are permitted provided that the following conditions are met: 6 | 7 | * Redistributions of source code must retain the above copyright notice, this 8 | list of conditions and the following disclaimer. 9 | 10 | * Redistributions in binary form must reproduce the above copyright notice, 11 | this list of conditions and the following disclaimer in the documentation 12 | and/or other materials provided with the distribution. 13 | 14 | THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" 15 | AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 16 | IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE 17 | DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE 18 | FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 19 | DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR 20 | SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER 21 | CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, 22 | OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE 23 | OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 24 | 25 | -------------------------------------------------------------------------------- /README.md: -------------------------------------------------------------------------------- 1 | # Citation 2 | Please cite 3 | http://arxiv.org/abs/1509.03371 4 | when using this repository for research. 5 | 6 | # Dataset licences 7 | 8 | ### When using dataset_01, please cite the following source: 9 | ``` 10 | @online{DS1ssTEM, 11 | title = {Segmented anisotropic ssTEM dataset of neural tissue}, 12 | author = {Gerhard, Stephan and Funke, Jan and Martel, Julien and Cardona, Albert and Fetter, Richard}, 13 | year = 2013, 14 | url = {http://dx.doi.org/10.6084/m9.figshare.856713}, 15 | doi = {10.6084/m9.figshare.856713}, 16 | urldate = {2015-08-20} 17 | } 18 | 19 | ``` 20 | 21 | ### When using dataset_02, please cite the following source: 22 | ``` 23 | @online{ISBI2012, 24 | title = {ISBI 2012 Challenge}, 25 | url = {http://brainiac2.mit.edu/isbi_challenge/}, 26 | urldate = {2015-08-20} 27 | } 28 | 29 | ``` 30 | and 31 | ``` 32 | @article{Cardona2010, 33 | doi = {10.1371/journal.pbio.1000502}, 34 | url = {http://dx.doi.org/10.1371/journal.pbio.1000502}, 35 | year = {2010}, 36 | month = {oct}, 37 | publisher = {Public Library of Science ({PLoS})}, 38 | volume = {8}, 39 | number = {10}, 40 | pages = {e1000502}, 41 | author = {Albert Cardona and Stephan Saalfeld and Stephan Preibisch and Benjamin Schmid and Anchi Cheng and Jim Pulokas and Pavel Tomancak and Volker Hartenstein}, 42 | editor = {Kristen M. Harris}, 43 | title = {An Integrated Micro- and Macroarchitectural Analysis of the Drosophila Brain by Computer-Assisted Serial Section Electron Microscopy}, 44 | journal = {{PLoS} Biology} 45 | } 46 | ``` 47 | and 48 | ``` 49 | @online{TrackEM, 50 | title = {TrackEM}, 51 | url = {https://www.ini.uzh.ch/~acardona/trakem2.html}, 52 | year = {2012}, 53 | urldate = {2015-08-20} 54 | } 55 | ``` 56 | -------------------------------------------------------------------------------- /benchmark/GTX980_CUDA/sk_benchmark/process_forward_layers_timings.csv: -------------------------------------------------------------------------------- 1 | 0;data;0.014043;0 2 | 1;silence;0.0018146;0 3 | 2;conv1;1.0959422;699388656 4 | 3;relu1;0.2368284;0 5 | 4;pool1;0.3240114;0 6 | 5;conv2;5.4014258;14062669312 7 | 6;relu2;0.318238;0 8 | 7;pool2;0.71559;0 9 | 8;conv3;18.321964;18401596416 10 | 9;relu3;0.4395918;0 11 | 10;pool3;1.0135224;0 12 | 11;ip1;154.3489266;644228317184 13 | 12;relu4;0.8643864;0 14 | 13;ip2;5.990901;17171480576 15 | 14;relu5;0.4132996;0 16 | 15;ip3;0.9644582;33521664 17 | 16;prob;0.047309;0 18 | -------------------------------------------------------------------------------- /benchmark/GTX980_CUDA/sk_benchmark/process_memory_usage.csv: -------------------------------------------------------------------------------- 1 | Peak memory usage;2892571592 2 | data;629292 3 | label;4 4 | conv1;9547968 5 | pool1;9462528 6 | conv2;23447552 7 | pool2;23011328 8 | conv3;31961088 9 | pool3;30720000 10 | ip1;67108864 11 | ip2;33554432 12 | ip3;131072 13 | prob;131072 14 | -------------------------------------------------------------------------------- /benchmark/GTX980_CUDA/sk_benchmark/process_total_timings.csv: -------------------------------------------------------------------------------- 1 | Forward;191.714908 2 | -------------------------------------------------------------------------------- /benchmark/GTX980_CUDA/sk_benchmark/train_backward_layers_timings.csv: -------------------------------------------------------------------------------- 1 | 0;label;0.000877;0 2 | 1;data;0.0009696;0 3 | 2;silence;0.001661;0 4 | 3;conv1;1.819903;0 5 | 4;relu1;0.2726896;0 6 | 5;pool1;0.6076484;0 7 | 6;conv2;17.9877856;0 8 | 7;relu2;0.4169074;0 9 | 8;pool2;1.3063268;0 10 | 9;conv3;18.9981138;0 11 | 10;relu3;0.6648252;0 12 | 11;pool3;1.907103;0 13 | 12;ip1;371.5397424;0 14 | 13;relu4;1.3922454;0 15 | 14;ip2;13.5833212;0 16 | 15;relu5;0.625952;0 17 | 16;ip3;2.4598556;0 18 | 17;loss;0.0460906;0 19 | -------------------------------------------------------------------------------- /benchmark/GTX980_CUDA/sk_benchmark/train_forward_layers_timings.csv: -------------------------------------------------------------------------------- 1 | 0;label;0.0187622;0 2 | 1;data;0.003997;0 3 | 2;silence;0.0019424;0 4 | 3;conv1;1.1397308;699388656 5 | 4;relu1;0.1666118;0 6 | 5;pool1;0.2980992;0 7 | 6;conv2;5.4553708;14062669312 8 | 7;relu2;0.303442;0 9 | 8;pool2;0.694105;0 10 | 9;conv3;17.3049408;18401596416 11 | 10;relu3;0.407844;0 12 | 11;pool3;0.944289;0 13 | 12;ip1;153.562362;644228317184 14 | 13;relu4;0.8710318;0 15 | 14;ip2;5.872702;17171480576 16 | 15;relu5;0.4279062;0 17 | 16;ip3;0.969711;33521664 18 | 17;loss;0.3509824;0 19 | -------------------------------------------------------------------------------- /benchmark/GTX980_CUDA/sk_benchmark/train_memory_usage.csv: -------------------------------------------------------------------------------- 1 | Peak memory usage;3204621764 2 | label;65536 3 | labeli;4 4 | data;629292 5 | datai;4 6 | conv1;9547968 7 | pool1;9462528 8 | conv2;23447552 9 | pool2;23011328 10 | conv3;31961088 11 | pool3;30720000 12 | ip1;67108864 13 | ip2;33554432 14 | ip3;131072 15 | (automatic);4 16 | -------------------------------------------------------------------------------- /benchmark/GTX980_CUDA/sk_benchmark/train_total_timings.csv: -------------------------------------------------------------------------------- 1 | Forward;189.6121536 2 | Backward;434.9790178 3 | -------------------------------------------------------------------------------- /benchmark/GTX980_CUDA/u_benchmark/process_forward_layers_timings.csv: -------------------------------------------------------------------------------- 1 | 0;data;0.0201474;0 2 | 1;silence;0.0018914;0 3 | 2;conv1;3.1065658;1102060800 4 | 3;relu1;1.0075428;0 5 | 4;conv2;11.7317112;23765774336 6 | 5;relu2;1.0244426;0 7 | 6;conv2_relu2_0_split;0.0041298;0 8 | 7;pool1;0.8950078;0 9 | 8;conv3;4.7744606;11716111872 10 | 9;relu3;0.525515;0 11 | 10;conv4;8.5238032;23111065600 12 | 11;relu4;0.5106686;0 13 | 12;conv4_relu4_0_split;0.0070484;0 14 | 13;pool2;0.4979998;0 15 | 14;conv5;3.8086208;11227732992 16 | 15;relu5;0.2423108;0 17 | 16;conv6;6.4169956;21814034432 18 | 17;relu6;0.301866;0 19 | 18;conv6_relu6_0_split;0.0033552;0 20 | 19;pool3;0.2101008;0 21 | 20;conv7;2.8681432;10274863104 22 | 21;relu7;0.1139792;0 23 | 22;conv8;4.7578372;19325255680 24 | 23;relu8;0.1661732;0 25 | 24;conv8_relu8_0_split;0.0037938;0 26 | 25;pool4;0.0895564;0 27 | 26;conv9;2.3970408;8492544000 28 | 27;relu9;0.0752706;0 29 | 28;conv10;3.8179296;14796701696 30 | 29;relu10;0.0696692;0 31 | 30;upconv1;4.6572626;0 32 | 31;conv11;0.9122462;3286728704 33 | 32;mergecrop1;0.4061404;0 34 | 33;conv12;8.9003274;27517335552 35 | 34;relu11;0.1295904;0 36 | 35;conv13;3.6501898;12757688320 37 | 36;relu12;0.07381;0 38 | 37;upconv2;3.1007358;0 39 | 38;conv14;0.863489;2832580608 40 | 39;mergecrop2;0.6454796;0 41 | 40;conv15;7.4727116;24543452160 42 | 41;relu13;0.1346504;0 43 | 42;conv16;3.4077856;11793920000 44 | 43;relu14;0.128637;0 45 | 44;upconv3;3.1867074;0 46 | 45;conv17;0.9284416;2616320000 47 | 46;mergecrop3;1.2322908;0 48 | 47;conv18;8.0772324;23118441984 49 | 48;relu15;0.2757638;0 50 | 49;conv19;4.1183232;11324422144 51 | 50;relu16;0.2417348;0 52 | 51;upconv4;4.7143024;0 53 | 52;conv20;1.365009;2507796480 54 | 53;mergecrop4;2.2628018;0 55 | 54;conv21;10.7722956;22418323200 56 | 55;relu17;0.472244;0 57 | 56;conv22;5.5281674;11089673216 58 | 57;relu18;0.493121;0 59 | 58;ip1;0.3460302;38238176 60 | 59;prob;0.1049962;0 61 | -------------------------------------------------------------------------------- /benchmark/GTX980_CUDA/u_benchmark/process_memory_usage.csv: -------------------------------------------------------------------------------- 1 | Peak memory usage;1793039512 2 | data;3926208 3 | datai;4 4 | conv1;83174400 5 | conv2;82591744 6 | conv2_relu2_0_split_0;82591744 7 | conv2_relu2_0_split_1;82591744 8 | pool1;20647936 9 | conv3;40716288 10 | conv4;40140800 11 | conv4_relu4_0_split_0;40140800 12 | conv4_relu4_0_split_1;40140800 13 | pool2;10035200 14 | conv5;19501056 15 | conv6;18939904 16 | conv6_relu6_0_split_0;18939904 17 | conv6_relu6_0_split_1;18939904 18 | pool3;4734976 19 | conv7;8921088 20 | conv8;8388608 21 | conv8_relu8_0_split_0;8388608 22 | conv8_relu8_0_split_1;8388608 23 | pool4;2097152 24 | conv9;3686400 25 | conv10;3211264 26 | upconv1;12845056 27 | conv11;6422528 28 | mergecrop1;12845056 29 | conv12;5971968 30 | conv13;5537792 31 | upconv2;22151168 32 | conv14;11075584 33 | mergecrop2;22151168 34 | conv15;10653696 35 | conv16;10240000 36 | upconv3;40960000 37 | conv17;20480000 38 | mergecrop3;40960000 39 | conv18;20072448 40 | conv19;19668992 41 | upconv4;78675968 42 | conv20;39337984 43 | mergecrop4;78675968 44 | conv21;38937600 45 | conv22;38539264 46 | ip1;1204352 47 | prob;1204352 48 | -------------------------------------------------------------------------------- /benchmark/GTX980_CUDA/u_benchmark/process_total_timings.csv: -------------------------------------------------------------------------------- 1 | Forward;142.274348 2 | -------------------------------------------------------------------------------- /benchmark/GTX980_CUDA/usk_benchmark/process_forward_layers_timings.csv: -------------------------------------------------------------------------------- 1 | 0;data;0.014555;0 2 | 1;silence;0.0013558;0 3 | 2;conv1;1.9568992;638903552 4 | 3;relu1;1.1153764;0 5 | 4;conv2;7.0789774;13747470336 6 | 5;relu2;0.6093588;0 7 | 6;conv2_relu2_0_split;0.0031992;0 8 | 7;pool1;0.4779596;0 9 | 8;conv3;9.0965702;26253855616 10 | 9;relu3;0.3188322;0 11 | 10;pool2;0.7617608;0 12 | 11;conv4;7.5546304;21813442560 13 | 12;relu4;0.351455;0 14 | 13;pool3;0.6009214;0 15 | 14;conv5;6.417378;18922176000 16 | 15;relu5;0.228116;0 17 | 16;pool4;0.5110046;0 18 | 17;ip1;35.4771948;141758822400 19 | 18;relu6;0.4278384;0 20 | 19;ip2;1.3201184;4425907200 21 | 20;relu7;0.2133016;0 22 | 21;upconv1;4.6721504;0 23 | 22;conv6;1.512341;4421580800 24 | 23;mergecrop1;1.529794;0 25 | 24;conv7;10.4137052;29437263360 26 | 25;relu8;0.4181806;0 27 | 26;conv8;4.5365524;9659482112 28 | 27;relu9;0.2285804;0 29 | 28;ip3;0.2084614;16646144 30 | 29;prob;0.064374;0 31 | -------------------------------------------------------------------------------- /benchmark/GTX980_CUDA/usk_benchmark/process_memory_usage.csv: -------------------------------------------------------------------------------- 1 | Peak memory usage;1145971772 2 | data;2281152 3 | datai;4 4 | conv1;48219136 5 | conv2;47775744 6 | conv2_relu2_0_split_0;47775744 7 | conv2_relu2_0_split_1;47775744 8 | pool1;11943936 9 | conv3;22794752 10 | pool2;22579200 11 | conv4;21307392 12 | pool3;20891648 13 | conv5;18483200 14 | pool4;17713152 15 | ip1;34611200 16 | ip2;17305600 17 | upconv1;69222400 18 | conv6;34611200 19 | mergecrop1;51916800 20 | conv7;34080768 21 | conv8;16777216 22 | ip3;524288 23 | prob;524288 24 | -------------------------------------------------------------------------------- /benchmark/GTX980_CUDA/usk_benchmark/process_total_timings.csv: -------------------------------------------------------------------------------- 1 | Forward;99.5959904 2 | -------------------------------------------------------------------------------- /benchmark/GTX980_OPENCL/sk_benchmark/process_forward_layers_timings.csv: -------------------------------------------------------------------------------- 1 | 0;data;0.0143252;0 2 | 1;silence;0.0026662;0 3 | 2;conv1;1.7259672;699388656 4 | 3;relu1;0.1370306;0 5 | 4;pool1;0.3360722;0 6 | 5;conv2;11.2391658;14062669312 7 | 6;relu2;0.3190846;0 8 | 7;pool2;0.7757316;0 9 | 8;conv3;18.0968498;18401596416 10 | 9;relu3;0.450221;0 11 | 10;pool3;1.051857;0 12 | 11;ip1;514.4112442;644228317184 13 | 12;relu4;0.9857154;0 14 | 13;ip2;15.6233562;17171480576 15 | 14;relu5;0.4618694;0 16 | 15;ip3;0.964134;33521664 17 | 16;prob;0.074961;0 18 | -------------------------------------------------------------------------------- /benchmark/GTX980_OPENCL/sk_benchmark/process_memory_usage.csv: -------------------------------------------------------------------------------- 1 | Peak memory usage;2892571592 2 | data;629292 3 | label;4 4 | conv1;9547968 5 | pool1;9462528 6 | conv2;23447552 7 | pool2;23011328 8 | conv3;31961088 9 | pool3;30720000 10 | ip1;67108864 11 | ip2;33554432 12 | ip3;131072 13 | prob;131072 14 | -------------------------------------------------------------------------------- /benchmark/GTX980_OPENCL/sk_benchmark/process_total_timings.csv: -------------------------------------------------------------------------------- 1 | Forward;575.669384 2 | -------------------------------------------------------------------------------- /benchmark/GTX980_OPENCL/sk_benchmark/train_backward_layers_timings.csv: -------------------------------------------------------------------------------- 1 | 0;label;0.0010642;0 2 | 1;data;0.0010558;0 3 | 2;silence;0.001359;0 4 | 3;conv1;4.1609056;0 5 | 4;relu1;0.2090056;0 6 | 5;pool1;0.5922258;0 7 | 6;conv2;33.3982292;0 8 | 7;relu2;0.4649138;0 9 | 8;pool2;1.380455;0 10 | 9;conv3;56.4363548;0 11 | 10;relu3;0.7740224;0 12 | 11;pool3;1.9889836;0 13 | 12;ip1;1708.289885;0 14 | 13;relu4;1.3190932;0 15 | 14;ip2;51.176079;0 16 | 15;relu5;0.6347298;0 17 | 16;ip3;2.7247558;0 18 | 17;loss;0.1924842;0 19 | -------------------------------------------------------------------------------- /benchmark/GTX980_OPENCL/sk_benchmark/train_forward_layers_timings.csv: -------------------------------------------------------------------------------- 1 | 0;label;0.0147486;0 2 | 1;data;0.005773;0 3 | 2;silence;0.0024298;0 4 | 3;conv1;1.7548638;699388656 5 | 4;relu1;0.1631776;0 6 | 5;pool1;0.3609746;0 7 | 6;conv2;11.474481;14062669312 8 | 7;relu2;0.3140282;0 9 | 8;pool2;0.7902454;0 10 | 9;conv3;18.5113284;18401596416 11 | 10;relu3;0.448919;0 12 | 11;pool3;1.045839;0 13 | 12;ip1;517.236149;644228317184 14 | 13;relu4;1.0337504;0 15 | 14;ip2;15.7873238;17171480576 16 | 15;relu5;0.4720762;0 17 | 16;ip3;0.9371168;33521664 18 | 17;loss;0.7269272;0 19 | -------------------------------------------------------------------------------- /benchmark/GTX980_OPENCL/sk_benchmark/train_memory_usage.csv: -------------------------------------------------------------------------------- 1 | Peak memory usage;3204621764 2 | label;65536 3 | labeli;4 4 | data;629292 5 | datai;4 6 | conv1;9547968 7 | pool1;9462528 8 | conv2;23447552 9 | pool2;23011328 10 | conv3;31961088 11 | pool3;30720000 12 | ip1;67108864 13 | ip2;33554432 14 | ip3;131072 15 | (automatic);4 16 | -------------------------------------------------------------------------------- /benchmark/GTX980_OPENCL/sk_benchmark/train_total_timings.csv: -------------------------------------------------------------------------------- 1 | Forward;565.7039322 2 | Backward;1864.502751 3 | -------------------------------------------------------------------------------- /benchmark/GTX980_OPENCL/u_benchmark/process_forward_layers_timings.csv: -------------------------------------------------------------------------------- 1 | 0;data;0.0187342;0 2 | 1;silence;0.0019274;0 3 | 2;conv1;3.6299318;1102060800 4 | 3;relu1;1.0869384;0 5 | 4;conv2;26.3692056;23765774336 6 | 5;relu2;1.1215754;0 7 | 6;conv2_relu2_0_split;0.0062658;0 8 | 7;pool1;0.8576278;0 9 | 8;conv3;9.7327258;11716111872 10 | 9;relu3;0.5292008;0 11 | 10;conv4;21.8746116;23111065600 12 | 11;relu4;0.5611102;0 13 | 12;conv4_relu4_0_split;0.0042962;0 14 | 13;pool2;0.4425734;0 15 | 14;conv5;8.1746674;11227732992 16 | 15;relu5;0.2973982;0 17 | 16;conv6;19.146688;21814034432 18 | 17;relu6;0.270644;0 19 | 18;conv6_relu6_0_split;0.0033782;0 20 | 19;pool3;0.2208972;0 21 | 20;conv7;6.894211;10274863104 22 | 21;relu7;0.1427362;0 23 | 22;conv8;16.823695;19325255680 24 | 23;relu8;0.1227904;0 25 | 24;conv8_relu8_0_split;0.003319;0 26 | 25;pool4;0.098882;0 27 | 26;conv9;5.5924726;8492544000 28 | 27;relu9;0.0790044;0 29 | 28;conv10;12.8098188;14796701696 30 | 29;relu10;0.0897474;0 31 | 30;upconv1;56.0273676;0 32 | 31;conv11;2.6819994;3286728704 33 | 32;mergecrop1;0.3769342;0 34 | 33;conv12;18.4179932;27517335552 35 | 34;relu11;0.0932134;0 36 | 35;conv13;11.5935164;12757688320 37 | 36;relu12;0.108353;0 38 | 37;upconv2;28.5042182;0 39 | 38;conv14;2.5558186;2832580608 40 | 39;mergecrop2;0.5757984;0 41 | 40;conv15;18.1934352;24543452160 42 | 41;relu13;0.1733488;0 43 | 42;conv16;10.6109398;11793920000 44 | 43;relu14;0.1419536;0 45 | 44;upconv3;15.3317548;0 46 | 45;conv17;2.6098348;2616320000 47 | 46;mergecrop3;1.0482962;0 48 | 47;conv18;18.019848;23118441984 49 | 48;relu15;0.2833876;0 50 | 49;conv19;11.3220628;11324422144 51 | 50;relu16;0.2591988;0 52 | 51;upconv4;9.6924594;0 53 | 52;conv20;2.848724;2507796480 54 | 53;mergecrop4;1.933022;0 55 | 54;conv21;20.446861;22418323200 56 | 55;relu17;0.5064346;0 57 | 56;conv22;12.763158;11089673216 58 | 57;relu18;0.53197;0 59 | 58;ip1;0.4512028;38238176 60 | 59;prob;0.128015;0 61 | -------------------------------------------------------------------------------- /benchmark/GTX980_OPENCL/u_benchmark/process_memory_usage.csv: -------------------------------------------------------------------------------- 1 | Peak memory usage;1793039512 2 | data;3926208 3 | datai;4 4 | conv1;83174400 5 | conv2;82591744 6 | conv2_relu2_0_split_0;82591744 7 | conv2_relu2_0_split_1;82591744 8 | pool1;20647936 9 | conv3;40716288 10 | conv4;40140800 11 | conv4_relu4_0_split_0;40140800 12 | conv4_relu4_0_split_1;40140800 13 | pool2;10035200 14 | conv5;19501056 15 | conv6;18939904 16 | conv6_relu6_0_split_0;18939904 17 | conv6_relu6_0_split_1;18939904 18 | pool3;4734976 19 | conv7;8921088 20 | conv8;8388608 21 | conv8_relu8_0_split_0;8388608 22 | conv8_relu8_0_split_1;8388608 23 | pool4;2097152 24 | conv9;3686400 25 | conv10;3211264 26 | upconv1;12845056 27 | conv11;6422528 28 | mergecrop1;12845056 29 | conv12;5971968 30 | conv13;5537792 31 | upconv2;22151168 32 | conv14;11075584 33 | mergecrop2;22151168 34 | conv15;10653696 35 | conv16;10240000 36 | upconv3;40960000 37 | conv17;20480000 38 | mergecrop3;40960000 39 | conv18;20072448 40 | conv19;19668992 41 | upconv4;78675968 42 | conv20;39337984 43 | mergecrop4;78675968 44 | conv21;38937600 45 | conv22;38539264 46 | ip1;1204352 47 | prob;1204352 48 | -------------------------------------------------------------------------------- /benchmark/GTX980_OPENCL/u_benchmark/process_total_timings.csv: -------------------------------------------------------------------------------- 1 | Forward;385.8560848 2 | -------------------------------------------------------------------------------- /benchmark/GTX980_OPENCL/usk_benchmark/process_forward_layers_timings.csv: -------------------------------------------------------------------------------- 1 | 0;data;0.013036;0 2 | 1;silence;0.001986;0 3 | 2;conv1;2.788269;638903552 4 | 3;relu1;0.8252844;0 5 | 4;conv2;15.4850384;13747470336 6 | 5;relu2;0.650384;0 7 | 6;conv2_relu2_0_split;0.004021;0 8 | 7;pool1;0.4877516;0 9 | 8;conv3;28.2640398;26253855616 10 | 9;relu3;0.3278726;0 11 | 10;pool2;0.7759524;0 12 | 11;conv4;21.7133158;21813442560 13 | 12;relu4;0.3312742;0 14 | 13;pool3;0.7008708;0 15 | 14;conv5;14.9449984;18922176000 16 | 15;relu5;0.2526546;0 17 | 16;pool4;0.5994888;0 18 | 17;ip1;98.3455104;141758822400 19 | 18;relu6;0.4822476;0 20 | 19;ip2;3.4107204;4425907200 21 | 20;relu7;0.3354066;0 22 | 21;upconv1;16.4908672;0 23 | 22;conv6;4.502969;4421580800 24 | 23;mergecrop1;1.3090008;0 25 | 24;conv7;22.8922092;29437263360 26 | 25;relu8;0.4673992;0 27 | 26;conv8;10.2436214;9659482112 28 | 27;relu9;0.2223656;0 29 | 28;ip3;0.2090416;16646144 30 | 29;prob;0.0858128;0 31 | -------------------------------------------------------------------------------- /benchmark/GTX980_OPENCL/usk_benchmark/process_memory_usage.csv: -------------------------------------------------------------------------------- 1 | Peak memory usage;1145971772 2 | data;2281152 3 | datai;4 4 | conv1;48219136 5 | conv2;47775744 6 | conv2_relu2_0_split_0;47775744 7 | conv2_relu2_0_split_1;47775744 8 | pool1;11943936 9 | conv3;22794752 10 | pool2;22579200 11 | conv4;21307392 12 | pool3;20891648 13 | conv5;18483200 14 | pool4;17713152 15 | ip1;34611200 16 | ip2;17305600 17 | upconv1;69222400 18 | conv6;34611200 19 | mergecrop1;51916800 20 | conv7;34080768 21 | conv8;16777216 22 | ip3;524288 23 | prob;524288 24 | -------------------------------------------------------------------------------- /benchmark/GTX980_OPENCL/usk_benchmark/process_total_timings.csv: -------------------------------------------------------------------------------- 1 | Forward;245.0145208 2 | -------------------------------------------------------------------------------- /benchmark/Old Benchmarks/E5-2697v3_OPENCL/sk_benchmark/process_forward_layers_timings.csv: -------------------------------------------------------------------------------- 1 | 0;data;0.0405226 2 | 1;silence;0.0063974 3 | 2;conv1;47.0204064 4 | 3;relu1;5.0779144 5 | 4;pool1;17.1705162 6 | 5;conv2;176.9591032 7 | 6;relu2;14.042627 8 | 7;pool2;44.3998096 9 | 8;conv3;238.911284 10 | 9;relu3;17.4494256 11 | 10;pool3;59.0547566 12 | 11;ip1;4168.215566 13 | 12;relu4;47.6171252 14 | 13;ip2;187.0531798 15 | 14;relu5;24.0963336 16 | 15;ip3;77.5087656 17 | 16;prob;4.877636 18 | -------------------------------------------------------------------------------- /benchmark/Old Benchmarks/E5-2697v3_OPENCL/sk_benchmark/process_memory_usage.csv: -------------------------------------------------------------------------------- 1 | Peak memory usage;1635212260 2 | -------------------------------------------------------------------------------- /benchmark/Old Benchmarks/E5-2697v3_OPENCL/sk_benchmark/process_total_timings.csv: -------------------------------------------------------------------------------- 1 | Forward;5161.034542 2 | -------------------------------------------------------------------------------- /benchmark/Old Benchmarks/E5-2697v3_OPENCL/sk_benchmark/train_backward_layers_timings.csv: -------------------------------------------------------------------------------- 1 | 0;label;0.006244 2 | 1;data;0.0173352 3 | 2;silence;0.0480516 4 | 3;conv1;27.407684 5 | 4;relu1;9.3572536 6 | 5;pool1;31.7626618 7 | 6;conv2;223.6084942 8 | 7;relu2;9.7248338 9 | 8;pool2;34.830752 10 | 9;conv3;259.413099 11 | 10;relu3;14.0322476 12 | 11;pool3;35.4135304 13 | 12;ip1;2187.657496 14 | 13;relu4;11.3234554 15 | 14;ip2;93.2401802 16 | 15;relu5;7.8386576 17 | 16;ip3;30.7950094 18 | 17;loss;1.0217236 19 | -------------------------------------------------------------------------------- /benchmark/Old Benchmarks/E5-2697v3_OPENCL/sk_benchmark/train_forward_layers_timings.csv: -------------------------------------------------------------------------------- 1 | 0;label;0.0351444 2 | 1;data;0.0188018 3 | 2;silence;0.0073194 4 | 3;conv1;19.201851 5 | 4;relu1;3.2664994 6 | 5;pool1;6.7968486 7 | 6;conv2;96.6467126 8 | 7;relu2;9.7047604 9 | 8;pool2;21.8509584 10 | 9;conv3;114.3279984 11 | 10;relu3;8.0902864 12 | 11;pool3;23.6304708 13 | 12;ip1;800.3068362 14 | 13;relu4;12.7776288 15 | 14;ip2;48.8153678 16 | 15;relu5;5.4744796 17 | 16;ip3;15.130366 18 | 17;loss;4.8494892 19 | -------------------------------------------------------------------------------- /benchmark/Old Benchmarks/E5-2697v3_OPENCL/sk_benchmark/train_memory_usage.csv: -------------------------------------------------------------------------------- 1 | Peak memory usage;999151612 2 | -------------------------------------------------------------------------------- /benchmark/Old Benchmarks/E5-2697v3_OPENCL/sk_benchmark/train_total_timings.csv: -------------------------------------------------------------------------------- 1 | Forward;1264.53591 2 | Backward;2532.156276 3 | -------------------------------------------------------------------------------- /benchmark/Old Benchmarks/E5-2697v3_OPENCL/u_benchmark/process_forward_layers_timings.csv: -------------------------------------------------------------------------------- 1 | 0;data;0.0198354 2 | 1;silence;0.003632 3 | 2;conv1;55.090086 4 | 3;relu1;41.0120502 5 | 4;conv2;1013.375996 6 | 5;relu2;41.3929282 7 | 6;conv2_relu2_0_split;0.0318762 8 | 7;pool1;51.4747408 9 | 8;conv3;206.1428936 10 | 9;relu3;23.7557262 11 | 10;conv4;442.6873478 12 | 11;relu4;24.5098422 13 | 12;conv4_relu4_0_split;0.0829166 14 | 13;pool2;34.2339488 15 | 14;conv5;119.1900464 16 | 15;relu5;11.457327 17 | 16;conv6;281.1877198 18 | 17;relu6;9.6852336 19 | 18;conv6_relu6_0_split;0.0223634 20 | 19;pool3;11.6210244 21 | 20;conv7;70.6778644 22 | 21;relu7;5.958447 23 | 22;conv8;145.9417566 24 | 23;relu8;4.2409396 25 | 24;conv8_relu8_0_split;0.0183822 26 | 25;pool4;7.111277 27 | 26;conv9;56.872058 28 | 27;relu9;2.3141562 29 | 28;conv10;96.5061614 30 | 29;relu10;2.3368562 31 | 30;upconv1;207.6817288 32 | 31;conv11;19.673262 33 | 32;mergecrop1;13.6522218 34 | 33;conv12;223.7458718 35 | 34;relu11;3.9363162 36 | 35;conv13;134.1608046 37 | 36;relu12;3.9666558 38 | 37;upconv2;141.2597508 39 | 38;conv14;27.783814 40 | 39;mergecrop2;31.2126118 41 | 40;conv15;312.4602684 42 | 41;relu13;8.95379 43 | 42;conv16;155.2613796 44 | 43;relu14;5.9838524 45 | 44;upconv3;130.4981226 46 | 45;conv17;24.607104 47 | 46;mergecrop3;57.296687 48 | 47;conv18;414.135743 49 | 48;relu15;13.025009 50 | 49;conv19;229.4715494 51 | 50;relu16;11.8842748 52 | 51;upconv4;164.088669 53 | 52;conv20;41.9184634 54 | 53;mergecrop4;111.9579278 55 | 54;conv21;714.7979378 56 | 55;relu17;19.549961 57 | 56;conv22;354.3634662 58 | 57;relu18;15.3219992 59 | 58;ip1;27.4159408 60 | 59;prob;16.2410806 61 | -------------------------------------------------------------------------------- /benchmark/Old Benchmarks/E5-2697v3_OPENCL/u_benchmark/process_memory_usage.csv: -------------------------------------------------------------------------------- 1 | Peak memory usage;1801471796 2 | -------------------------------------------------------------------------------- /benchmark/Old Benchmarks/E5-2697v3_OPENCL/u_benchmark/process_total_timings.csv: -------------------------------------------------------------------------------- 1 | Forward;6274.409446 2 | -------------------------------------------------------------------------------- /benchmark/Old Benchmarks/E5-2697v3_OPENCL/u_benchmark/train_backward_layers_timings.csv: -------------------------------------------------------------------------------- 1 | 0;label;0.010658 2 | 1;data;0.0512226 3 | 2;silence;0.0137026 4 | 3;conv1;86.4319616 5 | 4;relu1;57.2456308 6 | 5;conv2;1410.420551 7 | 6;relu2;76.032884 8 | 7;conv2_relu2_0_split;103.0249544 9 | 8;pool1;144.445817 10 | 9;conv3;430.3813386 11 | 10;relu3;30.6087382 12 | 11;conv4;772.0353558 13 | 12;relu4;32.632632 14 | 13;conv4_relu4_0_split;50.3196436 15 | 14;pool2;61.9868534 16 | 15;conv5;262.6665916 17 | 16;relu5;10.9531232 18 | 17;conv6;530.8880388 19 | 18;relu6;12.3195468 20 | 19;conv6_relu6_0_split;40.1771406 21 | 20;pool3;45.8929016 22 | 21;conv7;178.325765 23 | 22;relu7;6.9972068 24 | 23;conv8;333.2806924 25 | 24;relu8;18.8933454 26 | 25;conv8_relu8_0_split;17.1350212 27 | 26;pool4;32.671422 28 | 27;conv9;137.3401436 29 | 28;relu9;2.8333098 30 | 29;conv10;227.8165914 31 | 30;relu10;4.7558584 32 | 31;upconv1;200.0908172 33 | 32;conv11;36.8331184 34 | 33;mergecrop1;17.5279832 35 | 34;conv12;433.0395908 36 | 35;relu11;3.1266896 37 | 36;conv13;192.9265028 38 | 37;relu12;10.8964796 39 | 38;upconv2;131.869475 40 | 39;conv14;36.980428 41 | 40;mergecrop2;18.066299 42 | 41;conv15;455.7792268 43 | 42;relu13;4.8784866 44 | 43;conv16;218.8227134 45 | 44;relu14;10.9477738 46 | 45;upconv3;125.3655338 47 | 46;conv17;50.701712 48 | 47;mergecrop3;21.8793 49 | 48;conv18;774.5040862 50 | 49;relu15;12.9069906 51 | 50;conv19;388.760162 52 | 51;relu16;13.7737752 53 | 52;upconv4;146.8257822 54 | 53;conv20;77.952262 55 | 54;mergecrop4;54.6816494 56 | 55;conv21;1332.410706 57 | 56;relu17;38.0097954 58 | 57;conv22;614.4139046 59 | 58;relu18;40.2450944 60 | 59;ip1;19.4986164 61 | 60;loss;5.8554834 62 | -------------------------------------------------------------------------------- /benchmark/Old Benchmarks/E5-2697v3_OPENCL/u_benchmark/train_forward_layers_timings.csv: -------------------------------------------------------------------------------- 1 | 0;label;0.041263 2 | 1;data;0.0158122 3 | 2;silence;0.006677 4 | 3;conv1;49.1573852 5 | 4;relu1;51.0846154 6 | 5;conv2;859.2462698 7 | 6;relu2;61.6558684 8 | 7;conv2_relu2_0_split;0.111734 9 | 8;pool1;59.50288 10 | 9;conv3;216.2551428 11 | 10;relu3;30.3111696 12 | 11;conv4;484.4404536 13 | 12;relu4;31.4165364 14 | 13;conv4_relu4_0_split;0.0484288 15 | 14;pool2;28.7958146 16 | 15;conv5;120.3065616 17 | 16;relu5;13.4710082 18 | 17;conv6;242.4961242 19 | 18;relu6;15.1148608 20 | 19;conv6_relu6_0_split;0.119947 21 | 20;pool3;12.6655034 22 | 21;conv7;77.8326126 23 | 22;relu7;5.2534286 24 | 23;conv8;165.965359 25 | 24;relu8;5.1639906 26 | 25;conv8_relu8_0_split;0.063263 27 | 26;pool4;14.7202366 28 | 27;conv9;59.9360844 29 | 28;relu9;2.8721114 30 | 29;conv10;70.4706282 31 | 30;relu10;5.5082556 32 | 31;upconv1;186.0452722 33 | 32;conv11;19.8926912 34 | 33;mergecrop1;23.1774594 35 | 34;conv12;222.6629808 36 | 35;relu11;4.4860856 37 | 36;conv13;127.5499088 38 | 37;relu12;4.5859748 39 | 38;upconv2;153.0855106 40 | 39;conv14;22.1165958 41 | 40;mergecrop2;38.6550684 42 | 41;conv15;338.9844772 43 | 42;relu13;9.459336 44 | 43;conv16;175.3038108 45 | 44;relu14;6.575951 46 | 45;upconv3;157.69371 47 | 46;conv17;25.6478648 48 | 47;mergecrop3;71.255991 49 | 48;conv18;487.9113658 50 | 49;relu15;15.2723278 51 | 50;conv19;263.1543588 52 | 51;relu16;13.7430294 53 | 52;upconv4;169.978962 54 | 53;conv20;41.291798 55 | 54;mergecrop4;106.3960474 56 | 55;conv21;641.164679 57 | 56;relu17;24.93105 58 | 57;conv22;357.3861596 59 | 58;relu18;20.4734824 60 | 59;ip1;27.1241994 61 | 60;loss;24.0404162 62 | -------------------------------------------------------------------------------- /benchmark/Old Benchmarks/E5-2697v3_OPENCL/u_benchmark/train_memory_usage.csv: -------------------------------------------------------------------------------- 1 | Peak memory usage;3115754464 2 | -------------------------------------------------------------------------------- /benchmark/Old Benchmarks/E5-2697v3_OPENCL/u_benchmark/train_total_timings.csv: -------------------------------------------------------------------------------- 1 | Forward;6200.191189 2 | Backward;10730.3282 3 | -------------------------------------------------------------------------------- /benchmark/Old Benchmarks/E5-2697v3_OPENCL/usk_benchmark/process_forward_layers_timings.csv: -------------------------------------------------------------------------------- 1 | 0;data;0.026722 2 | 1;silence;0.0040646 3 | 2;conv1;80.3438366 4 | 3;relu1;68.5769352 5 | 4;conv2;960.6551128 6 | 5;relu2;49.7611038 7 | 6;conv2_relu2_0_split;0.0382458 8 | 7;pool1;78.4455622 9 | 8;conv3;227.0954262 10 | 9;relu3;20.720295 11 | 10;conv4;564.652915 12 | 11;relu4;34.5401934 13 | 12;conv4_relu4_0_split;0.027923 14 | 13;pool2;38.5501236 15 | 14;conv5;307.7153198 16 | 15;relu5;8.1091694 17 | 16;pool3;25.3235492 18 | 17;conv6;133.5515302 19 | 18;relu6;9.4014342 20 | 19;pool4;19.9955002 21 | 20;conv7;145.5263772 22 | 21;relu7;13.368182 23 | 22;pool5;44.2476982 24 | 23;ip1;221.3577124 25 | 24;relu8;22.5745366 26 | 25;ip2;29.9557704 27 | 26;relu9;13.1264358 28 | 27;upconv1;227.505374 29 | 28;conv8;33.6822922 30 | 29;mergecrop1;90.1352616 31 | 30;conv9;608.5342514 32 | 31;relu10;16.0964672 33 | 32;conv10;367.5179906 34 | 33;relu11;19.440321 35 | 34;upconv2;265.5485826 36 | 35;conv11;46.0486022 37 | 36;mergecrop2;149.6982854 38 | 37;conv12;840.1673394 39 | 38;relu12;49.614964 40 | 39;conv13;640.1183464 41 | 40;relu13;42.649099 42 | 41;ip3;64.2265004 43 | 42;prob;29.0756722 44 | -------------------------------------------------------------------------------- /benchmark/Old Benchmarks/E5-2697v3_OPENCL/usk_benchmark/process_memory_usage.csv: -------------------------------------------------------------------------------- 1 | Peak memory usage;3856535060 2 | -------------------------------------------------------------------------------- /benchmark/Old Benchmarks/E5-2697v3_OPENCL/usk_benchmark/process_total_timings.csv: -------------------------------------------------------------------------------- 1 | Forward;6623.801909 2 | -------------------------------------------------------------------------------- /benchmark/Old Benchmarks/E5-2697v3_OPENCL/usk_benchmark/train_backward_layers_timings.csv: -------------------------------------------------------------------------------- 1 | 0;label;0.0054054 2 | 1;data;0.0044142 3 | 2;silence;0.0056992 4 | 3;conv1;158.4993818 5 | 4;relu1;103.0751298 6 | 5;conv2;1873.491912 7 | 6;relu2;75.8166538 8 | 7;conv2_relu2_0_split;110.0896594 9 | 8;pool1;202.7994436 10 | 9;conv3;583.428602 11 | 10;relu3;40.1394528 12 | 11;conv4;979.2941216 13 | 12;relu4;54.1054634 14 | 13;conv4_relu4_0_split;106.5275346 15 | 14;pool2;86.4919004 16 | 15;conv5;620.510135 17 | 16;relu5;11.7958396 18 | 17;pool3;74.7497702 19 | 18;conv6;284.1231104 20 | 19;relu6;6.4438208 21 | 20;pool4;54.6364058 22 | 21;conv7;332.7911222 23 | 22;relu7;19.8424472 24 | 23;pool5;72.6593604 25 | 24;ip1;598.1666352 26 | 25;relu8;32.340552 27 | 26;ip2;59.9850716 28 | 27;relu9;13.4173976 29 | 28;upconv1;165.0862984 30 | 29;conv8;78.1315778 31 | 30;mergecrop1;103.3357076 32 | 31;conv9;1265.967189 33 | 32;relu10;24.4232914 34 | 33;conv10;634.9428754 35 | 34;relu11;30.9922038 36 | 35;upconv2;230.0025022 37 | 36;conv11;156.0840678 38 | 37;mergecrop2;465.0996004 39 | 38;conv12;2103.526447 40 | 39;relu12;51.7798402 41 | 40;conv13;1033.734255 42 | 41;relu13;65.559875 43 | 42;ip3;58.2909766 44 | 43;loss;7.1196846 45 | -------------------------------------------------------------------------------- /benchmark/Old Benchmarks/E5-2697v3_OPENCL/usk_benchmark/train_forward_layers_timings.csv: -------------------------------------------------------------------------------- 1 | 0;label;0.034628 2 | 1;data;0.0132282 3 | 2;silence;0.005266 4 | 3;conv1;71.8050074 5 | 4;relu1;67.7657314 6 | 5;conv2;1023.540374 7 | 6;relu2;69.3732522 8 | 7;conv2_relu2_0_split;0.0503286 9 | 8;pool1;86.531906 10 | 9;conv3;239.4939804 11 | 10;relu3;25.518366 12 | 11;conv4;516.9120138 13 | 12;relu4;27.8296604 14 | 13;conv4_relu4_0_split;0.0253668 15 | 14;pool2;43.3069702 16 | 15;conv5;275.2488314 17 | 16;relu5;10.5483852 18 | 17;pool3;24.5423482 19 | 18;conv6;133.0760446 20 | 19;relu6;9.80413 21 | 20;pool4;24.8255454 22 | 21;conv7;149.4610976 23 | 22;relu7;12.6427508 24 | 23;pool5;41.8787094 25 | 24;ip1;259.4693626 26 | 25;relu8;15.8684332 27 | 26;ip2;25.9545292 28 | 27;relu9;10.6594786 29 | 28;upconv1;255.0431336 30 | 29;conv8;36.6369454 31 | 30;mergecrop1;122.1754714 32 | 31;conv9;674.2189658 33 | 32;relu10;27.7444354 34 | 33;conv10;344.3291066 35 | 34;relu11;18.3370226 36 | 35;upconv2;225.357483 37 | 36;conv11;59.115188 38 | 37;mergecrop2;203.2147386 39 | 38;conv12;1041.086768 40 | 39;relu12;36.201827 41 | 40;conv13;644.1519768 42 | 41;relu13;31.6411922 43 | 42;ip3;59.4689534 44 | 43;loss;36.1787196 45 | -------------------------------------------------------------------------------- /benchmark/Old Benchmarks/E5-2697v3_OPENCL/usk_benchmark/train_memory_usage.csv: -------------------------------------------------------------------------------- 1 | Peak memory usage;5534664192 2 | -------------------------------------------------------------------------------- /benchmark/Old Benchmarks/E5-2697v3_OPENCL/usk_benchmark/train_total_timings.csv: -------------------------------------------------------------------------------- 1 | Forward;6643.204238 2 | Backward;13490.2204 3 | -------------------------------------------------------------------------------- /benchmark/Old Benchmarks/GTX980_CUDA/sk_benchmark/process_forward_layers_timings.csv: -------------------------------------------------------------------------------- 1 | 0;data;0.0048065 2 | 1;silence;0.00112395 3 | 2;conv1;1.05733825 4 | 3;relu1;0.1198611 5 | 4;pool1;0.2970089 6 | 5;conv2;5.1457242 7 | 6;relu2;0.28440505 8 | 7;pool2;0.70590725 9 | 8;conv3;16.86159505 10 | 9;relu3;0.38431225 11 | 10;pool3;0.9382434 12 | 11;ip1;158.9354779 13 | 12;relu4;0.8045739 14 | 13;ip2;5.95972205 15 | 14;relu5;0.407611 16 | 15;ip3;0.9841417 17 | 16;prob;0.05645365 18 | -------------------------------------------------------------------------------- /benchmark/Old Benchmarks/GTX980_CUDA/sk_benchmark/process_memory_usage.csv: -------------------------------------------------------------------------------- 1 | Peak memory usage;1635212260 2 | -------------------------------------------------------------------------------- /benchmark/Old Benchmarks/GTX980_CUDA/sk_benchmark/process_total_timings.csv: -------------------------------------------------------------------------------- 1 | Forward;193.2351756 2 | -------------------------------------------------------------------------------- /benchmark/Old Benchmarks/GTX980_CUDA/sk_benchmark/train_backward_layers_timings.csv: -------------------------------------------------------------------------------- 1 | 0;label;0.00080485 2 | 1;data;0.00081685 3 | 2;silence;0.0009576 4 | 3;conv1;0.91463575 5 | 4;relu1;0.08824025 6 | 5;pool1;0.28245015 7 | 6;conv2;7.47387405 8 | 7;relu2;0.2069305 9 | 8;pool2;0.6348041 10 | 9;conv3;6.60553235 11 | 10;relu3;0.26699105 12 | 11;pool3;0.80848305 13 | 12;ip1;77.663499 14 | 13;relu4;0.2992026 15 | 14;ip2;2.97334415 16 | 15;relu5;0.1532881 17 | 16;ip3;0.68998165 18 | 17;loss;0.0427047 19 | -------------------------------------------------------------------------------- /benchmark/Old Benchmarks/GTX980_CUDA/sk_benchmark/train_forward_layers_timings.csv: -------------------------------------------------------------------------------- 1 | 0;label;0.0044734 2 | 1;data;0.0015361 3 | 2;silence;0.00121455 4 | 3;conv1;0.5642506 5 | 4;relu1;0.0644302 6 | 5;pool1;0.15264065 7 | 6;conv2;2.5071563 8 | 7;relu2;0.14192065 9 | 8;pool2;0.33952445 10 | 9;conv3;7.73923715 11 | 10;relu3;0.18358155 12 | 11;pool3;0.4274263 13 | 12;ip1;39.33000855 14 | 13;relu4;0.20634705 15 | 14;ip2;1.46659905 16 | 15;relu5;0.1043175 17 | 16;ip3;0.2859803 18 | 17;loss;0.1219378 19 | -------------------------------------------------------------------------------- /benchmark/Old Benchmarks/GTX980_CUDA/sk_benchmark/train_memory_usage.csv: -------------------------------------------------------------------------------- 1 | Peak memory usage;999151612 2 | -------------------------------------------------------------------------------- /benchmark/Old Benchmarks/GTX980_CUDA/sk_benchmark/train_total_timings.csv: -------------------------------------------------------------------------------- 1 | Forward;53.5501298 2 | Backward;99.7280947 3 | -------------------------------------------------------------------------------- /benchmark/Old Benchmarks/GTX980_CUDA/u_benchmark/process_forward_layers_timings.csv: -------------------------------------------------------------------------------- 1 | 0;data;0.00722975 2 | 1;silence;0.00106755 3 | 2;conv1;2.92452625 4 | 3;relu1;1.02378065 5 | 4;conv2;11.5030544 6 | 5;relu2;0.98012145 7 | 6;conv2_relu2_0_split;0.0014339 8 | 7;pool1;0.7371831 9 | 8;conv3;4.5110732 10 | 9;relu3;0.4882995 11 | 10;conv4;8.165651 12 | 11;relu4;0.48023745 13 | 12;conv4_relu4_0_split;0.0011669 14 | 13;pool2;0.36415075 15 | 14;conv5;3.52254895 16 | 15;relu5;0.23777115 17 | 16;conv6;6.3480313 18 | 17;relu6;0.2295019 19 | 18;conv6_relu6_0_split;0.00114605 20 | 19;pool3;0.17735735 21 | 20;conv7;2.883017 22 | 21;relu7;0.11304685 23 | 22;conv8;4.8231356 24 | 23;relu8;0.10425335 25 | 24;conv8_relu8_0_split;0.0010977 26 | 25;pool4;0.08511455 27 | 26;conv9;2.3377986 28 | 27;relu9;0.04933965 29 | 28;conv10;3.89789365 30 | 29;relu10;0.0432682 31 | 30;upconv1;4.4088567 32 | 31;conv11;0.9223566 33 | 32;mergecrop1;0.3733388 34 | 33;conv12;8.94031225 35 | 34;relu11;0.07990545 36 | 35;conv13;3.594437 37 | 36;relu12;0.07096585 38 | 37;upconv2;3.07012515 39 | 38;conv14;0.86699635 40 | 39;mergecrop2;0.65308465 41 | 40;conv15;7.24313925 42 | 41;relu13;0.1371413 43 | 42;conv16;3.44476735 44 | 43;relu14;0.1258016 45 | 44;upconv3;3.12528635 46 | 45;conv17;0.9004486 47 | 46;mergecrop3;1.16783255 48 | 47;conv18;8.0232101 49 | 48;relu15;0.2488479 50 | 49;conv19;4.12113415 51 | 50;relu16;0.2374819 52 | 51;upconv4;4.66606145 53 | 52;conv20;1.3076 54 | 53;mergecrop4;2.23760385 55 | 54;conv21;10.3641045 56 | 55;relu17;0.46589955 57 | 56;conv22;5.50046915 58 | 57;relu18;0.45995625 59 | 58;ip1;0.39595055 60 | 59;prob;0.36968585 61 | -------------------------------------------------------------------------------- /benchmark/Old Benchmarks/GTX980_CUDA/u_benchmark/process_memory_usage.csv: -------------------------------------------------------------------------------- 1 | Peak memory usage;1801471796 2 | -------------------------------------------------------------------------------- /benchmark/Old Benchmarks/GTX980_CUDA/u_benchmark/process_total_timings.csv: -------------------------------------------------------------------------------- 1 | Forward;139.226524 2 | -------------------------------------------------------------------------------- /benchmark/Old Benchmarks/GTX980_CUDA/usk_benchmark/process_forward_layers_timings.csv: -------------------------------------------------------------------------------- 1 | 0;data;0.0080531 2 | 1;silence;0.00151975 3 | 2;conv1;4.34875615 4 | 3;relu1;1.4627367 5 | 4;conv2;16.9454736 6 | 5;relu2;1.44443605 7 | 6;conv2_relu2_0_split;0.0015194 8 | 7;pool1;1.107246 9 | 8;conv3;6.58756165 10 | 9;relu3;0.7256629 11 | 10;conv4;12.0125901 12 | 11;relu4;0.71291195 13 | 12;conv4_relu4_0_split;0.00113605 14 | 13;pool2;0.5368212 15 | 14;conv5;7.5439757 16 | 15;relu5;0.17421265 17 | 16;pool3;0.40789405 18 | 17;conv6;2.7882711 19 | 18;relu6;0.1623925 20 | 19;pool4;0.3777447 21 | 20;conv7;4.12992085 22 | 21;relu7;0.2848013 23 | 22;pool5;0.6386053 24 | 23;ip1;10.5724605 25 | 24;relu8;0.4220721 26 | 25;ip2;1.34537615 27 | 26;relu9;0.2148235 28 | 27;upconv1;4.76677375 29 | 28;conv8;1.56142555 30 | 29;mergecrop1;2.0029583 31 | 30;conv9;13.6182914 32 | 31;relu10;0.4152998 33 | 32;conv10;7.01692575 34 | 33;relu11;0.4109815 35 | 34;upconv2;7.80584785 36 | 35;conv11;2.14304255 37 | 36;mergecrop2;3.87495985 38 | 37;conv12;17.87997645 39 | 38;relu12;0.80659195 40 | 39;conv13;9.2228774 41 | 40;relu13;0.8024036 42 | 41;ip3;0.65446735 43 | 42;prob;0.69670635 44 | -------------------------------------------------------------------------------- /benchmark/Old Benchmarks/GTX980_CUDA/usk_benchmark/process_memory_usage.csv: -------------------------------------------------------------------------------- 1 | Peak memory usage;2639119892 2 | -------------------------------------------------------------------------------- /benchmark/Old Benchmarks/GTX980_CUDA/usk_benchmark/process_total_timings.csv: -------------------------------------------------------------------------------- 1 | Forward;150.0740029 2 | -------------------------------------------------------------------------------- /benchmark/Old Benchmarks/GTX980_OPENCL/sk_benchmark/process_forward_layers_timings.csv: -------------------------------------------------------------------------------- 1 | 0;data;0.006974 2 | 1;silence;0.0018588 3 | 2;conv1;1.58532795 4 | 3;relu1;0.12720155 5 | 4;pool1;0.31376995 6 | 5;conv2;11.8369024 7 | 6;relu2;0.3009139 8 | 7;pool2;0.7392743 9 | 8;conv3;15.989715 10 | 9;relu3;0.40724635 11 | 10;pool3;0.9777164 12 | 11;ip1;507.072725 13 | 12;relu4;0.8633914 14 | 13;ip2;15.13994745 15 | 14;relu5;0.4301443 16 | 15;ip3;1.143828 17 | 16;prob;0.0520418 18 | -------------------------------------------------------------------------------- /benchmark/Old Benchmarks/GTX980_OPENCL/sk_benchmark/process_memory_usage.csv: -------------------------------------------------------------------------------- 1 | Peak memory usage;1635212260 2 | -------------------------------------------------------------------------------- /benchmark/Old Benchmarks/GTX980_OPENCL/sk_benchmark/process_total_timings.csv: -------------------------------------------------------------------------------- 1 | Forward;557.7745539 2 | -------------------------------------------------------------------------------- /benchmark/Old Benchmarks/GTX980_OPENCL/u_benchmark/process_forward_layers_timings.csv: -------------------------------------------------------------------------------- 1 | 0;data;0.00955865 2 | 1;silence;0.0015038 3 | 2;conv1;3.2846843 4 | 3;relu1;1.0860999 5 | 4;conv2;25.45977305 6 | 5;relu2;1.07262275 7 | 6;conv2_relu2_0_split;0.0015992 8 | 7;pool1;0.8317198 9 | 8;conv3;10.39354905 10 | 9;relu3;0.5227294 11 | 10;conv4;21.135221 12 | 11;relu4;0.51264815 13 | 12;conv4_relu4_0_split;0.00115525 14 | 13;pool2;0.4138343 15 | 14;conv5;9.1477145 16 | 15;relu5;0.24891525 17 | 16;conv6;18.9615431 18 | 17;relu6;0.24150645 19 | 18;conv6_relu6_0_split;0.0012145 20 | 19;pool3;0.2048774 21 | 20;conv7;7.5637928 22 | 21;relu7;0.1186947 23 | 22;conv8;16.2464662 24 | 23;relu8;0.1111898 25 | 24;conv8_relu8_0_split;0.00124295 26 | 25;pool4;0.0997064 27 | 26;conv9;6.93520585 28 | 27;relu9;0.0537901 29 | 28;conv10;11.5031496 30 | 29;relu10;0.0478999 31 | 30;upconv1;51.6951237 32 | 31;conv11;2.8457067 33 | 32;mergecrop1;0.32431895 34 | 33;conv12;19.80052775 35 | 34;relu11;0.08229545 36 | 35;conv13;11.17782255 37 | 36;relu12;0.0761492 38 | 37;upconv2;26.4208572 39 | 38;conv14;2.565181 40 | 39;mergecrop2;0.54392685 41 | 40;conv15;18.9858533 42 | 41;relu13;0.14221895 43 | 42;conv16;9.8546206 44 | 43;relu14;0.1347123 45 | 44;upconv3;14.24252075 46 | 45;conv17;2.53708945 47 | 46;mergecrop3;0.98059345 48 | 47;conv18;19.4120579 49 | 48;relu15;0.2552756 50 | 49;conv19;10.2046753 51 | 50;relu16;0.25277825 52 | 51;upconv4;9.14812025 53 | 52;conv20;2.7453997 54 | 53;mergecrop4;1.8519993 55 | 54;conv21;21.8344965 56 | 55;relu17;0.49937505 57 | 56;conv22;11.7564781 58 | 57;relu18;0.48987325 59 | 58;ip1;0.59549685 60 | 59;prob;0.38081535 61 | -------------------------------------------------------------------------------- /benchmark/Old Benchmarks/GTX980_OPENCL/u_benchmark/process_memory_usage.csv: -------------------------------------------------------------------------------- 1 | Peak memory usage;1801471796 2 | -------------------------------------------------------------------------------- /benchmark/Old Benchmarks/GTX980_OPENCL/u_benchmark/process_total_timings.csv: -------------------------------------------------------------------------------- 1 | Forward;377.9963313 2 | -------------------------------------------------------------------------------- /benchmark/Old Benchmarks/GTX980_OPENCL/usk_benchmark/process_forward_layers_timings.csv: -------------------------------------------------------------------------------- 1 | 0;data;0.01613795 2 | 1;silence;0.00155345 3 | 2;conv1;4.90511305 4 | 3;relu1;1.5980401 5 | 4;conv2;37.54494645 6 | 5;relu2;1.58235515 7 | 6;conv2_relu2_0_split;0.0016131 8 | 7;pool1;1.2058553 9 | 8;conv3;15.4718482 10 | 9;relu3;0.77666865 11 | 10;conv4;30.72244695 12 | 11;relu4;0.765241 13 | 12;conv4_relu4_0_split;0.00119435 14 | 13;pool2;0.60044905 15 | 14;conv5;18.42941805 16 | 15;relu5;0.1816358 17 | 16;pool3;0.45380895 18 | 17;conv6;9.29531815 19 | 18;relu6;0.17200145 20 | 19;pool4;0.420933 21 | 20;conv7;15.5957153 22 | 21;relu7;0.30174775 23 | 22;pool5;0.71868975 24 | 23;ip1;43.34003 25 | 24;relu8;0.4509504 26 | 25;ip2;5.46416445 27 | 26;relu9;0.2264364 28 | 27;upconv1;15.40277645 29 | 28;conv8;3.9882935 30 | 29;mergecrop1;1.6495284 31 | 30;conv9;34.9728959 32 | 31;relu10;0.44555605 33 | 32;conv10;16.8238284 34 | 33;relu11;0.43396165 35 | 34;upconv2;11.22470735 36 | 35;conv11;4.68733925 37 | 36;mergecrop2;3.18047465 38 | 37;conv12;38.3764037 39 | 38;relu12;0.87239575 40 | 39;conv13;20.6236679 41 | 40;relu13;0.871285 42 | 41;ip3;0.9307859 43 | 42;prob;0.7180643 44 | -------------------------------------------------------------------------------- /benchmark/Old Benchmarks/GTX980_OPENCL/usk_benchmark/process_memory_usage.csv: -------------------------------------------------------------------------------- 1 | Peak memory usage;2639119892 2 | -------------------------------------------------------------------------------- /benchmark/Old Benchmarks/GTX980_OPENCL/usk_benchmark/process_total_timings.csv: -------------------------------------------------------------------------------- 1 | Forward;347.4728686 2 | -------------------------------------------------------------------------------- /benchmark/Old Benchmarks/W9100_OPENCL/sk_benchmark/process_forward_layers_timings.csv: -------------------------------------------------------------------------------- 1 | 0;data;0.06572795 2 | 1;silence;0.00436855 3 | 2;conv1;1.48202745 4 | 3;relu1;0.2582971 5 | 4;pool1;0.56307915 6 | 5;conv2;10.49666245 7 | 6;relu2;0.39177495 8 | 7;pool2;1.2589347 9 | 8;conv3;9.19528075 10 | 9;relu3;0.47658685 11 | 10;pool3;1.69524235 12 | 11;ip1;261.3687671 13 | 12;relu4;0.9276382 14 | 13;ip2;10.7197828 15 | 14;relu5;0.4916141 16 | 15;ip3;1.55598675 17 | 16;prob;0.15787245 18 | -------------------------------------------------------------------------------- /benchmark/Old Benchmarks/W9100_OPENCL/sk_benchmark/process_memory_usage.csv: -------------------------------------------------------------------------------- 1 | Peak memory usage;1635212260 2 | -------------------------------------------------------------------------------- /benchmark/Old Benchmarks/W9100_OPENCL/sk_benchmark/process_total_timings.csv: -------------------------------------------------------------------------------- 1 | Forward;301.0691711 2 | -------------------------------------------------------------------------------- /benchmark/Old Benchmarks/W9100_OPENCL/u_benchmark/process_forward_layers_timings.csv: -------------------------------------------------------------------------------- 1 | 0;data;0.0201097 2 | 1;silence;0.00619475 3 | 2;conv1;11.883502 4 | 3;relu1;1.3273735 5 | 4;conv2;28.73307785 6 | 5;relu2;1.119078 7 | 6;conv2_relu2_0_split;0.00491215 8 | 7;pool1;1.5388402 9 | 8;conv3;11.6333884 10 | 9;relu3;0.5118755 11 | 10;conv4;18.6725172 12 | 11;relu4;0.50147125 13 | 12;conv4_relu4_0_split;0.00438375 14 | 13;pool2;0.675312 15 | 14;conv5;7.9506434 16 | 15;relu5;0.26409705 17 | 16;conv6;14.180703 18 | 17;relu6;0.25455495 19 | 18;conv6_relu6_0_split;0.00564495 20 | 19;pool3;0.32355715 21 | 20;conv7;8.60981555 22 | 21;relu7;0.14543255 23 | 22;conv8;15.2585897 24 | 23;relu8;0.14985015 25 | 24;conv8_relu8_0_split;0.00808505 26 | 25;pool4;0.1720678 27 | 26;conv9;6.53874805 28 | 27;relu9;0.0962066 29 | 28;conv10;11.0678053 30 | 29;relu10;0.08994195 31 | 30;upconv1;15.0772938 32 | 31;conv11;2.8987925 33 | 32;mergecrop1;0.7113469 34 | 33;conv12;26.3816717 35 | 34;relu11;0.1471818 36 | 35;conv13;12.37683395 37 | 36;relu12;0.11754655 38 | 37;upconv2;8.7728592 39 | 38;conv14;2.69918305 40 | 39;mergecrop2;1.09742955 41 | 40;conv15;17.5822143 42 | 41;relu13;0.1633176 43 | 42;conv16;8.69196375 44 | 43;relu14;0.1562593 45 | 44;upconv3;6.94481235 46 | 45;conv17;2.48984455 47 | 46;mergecrop3;2.25167435 48 | 47;conv18;18.7959853 49 | 48;relu15;0.26578555 50 | 49;conv19;9.44985105 51 | 50;relu16;0.2583922 52 | 51;upconv4;9.6929859 53 | 52;conv20;2.40385315 54 | 53;mergecrop4;4.1987096 55 | 54;conv21;20.4765099 56 | 55;relu17;0.4744221 57 | 56;conv22;10.50956265 58 | 57;relu18;0.4657705 59 | 58;ip1;1.11723 60 | 59;prob;0.54379175 61 | -------------------------------------------------------------------------------- /benchmark/Old Benchmarks/W9100_OPENCL/u_benchmark/process_memory_usage.csv: -------------------------------------------------------------------------------- 1 | Peak memory usage;3115754464 2 | -------------------------------------------------------------------------------- /benchmark/Old Benchmarks/W9100_OPENCL/u_benchmark/process_total_timings.csv: -------------------------------------------------------------------------------- 1 | Forward;333.3243316 2 | -------------------------------------------------------------------------------- /benchmark/Old Benchmarks/W9100_OPENCL/u_benchmark/train_backward_layers_timings.csv: -------------------------------------------------------------------------------- 1 | 0;label;0.0112388 2 | 1;data;0.02455285 3 | 2;silence;0.02610325 4 | 3;conv1;22.90231645 5 | 4;relu1;1.260725 6 | 5;conv2;100.5647391 7 | 6;relu2;1.415693 8 | 7;conv2_relu2_0_split;1.33981385 9 | 8;pool1;5.47384695 10 | 9;conv3;33.2500064 11 | 10;relu3;0.6716867 12 | 11;conv4;38.4404541 13 | 12;relu4;0.7482094 14 | 13;conv4_relu4_0_split;0.6832467 15 | 14;pool2;2.63698495 16 | 15;conv5;19.49004765 17 | 16;relu5;0.3251737 18 | 17;conv6;21.7810867 19 | 18;relu6;0.40417525 20 | 19;conv6_relu6_0_split;0.3352051 21 | 20;pool3;1.2897005 22 | 21;conv7;14.17582705 23 | 22;relu7;0.16499545 24 | 23;conv8;21.93889155 25 | 24;relu8;0.2314259 26 | 25;conv8_relu8_0_split;0.16969535 27 | 26;pool4;0.59654295 28 | 27;conv9;11.670853 29 | 28;relu9;0.11606135 30 | 29;conv10;11.4172953 31 | 30;relu10;0.3078381 32 | 31;upconv1;16.0486073 33 | 32;conv11;3.72494655 34 | 33;mergecrop1;0.45276505 35 | 34;conv12;35.5666003 36 | 35;relu11;0.1517992 37 | 36;conv13;11.6264932 38 | 37;relu12;0.1828511 39 | 38;upconv2;8.5762882 40 | 39;conv14;4.58772965 41 | 40;mergecrop2;0.72157565 42 | 41;conv15;41.18420265 43 | 42;relu13;0.20215445 44 | 43;conv16;12.08773795 45 | 44;relu14;0.2306936 46 | 45;upconv3;5.8218835 47 | 46;conv17;10.9189093 48 | 47;mergecrop3;1.3285062 49 | 48;conv18;62.7123988 50 | 49;relu15;0.3512383 51 | 50;conv19;18.8257936 52 | 51;relu16;0.35032985 53 | 52;upconv4;6.06655275 54 | 53;conv20;35.2669816 55 | 54;mergecrop4;2.38453985 56 | 55;conv21;82.6659123 57 | 56;relu17;0.58634395 58 | 57;conv22;47.2391582 59 | 58;relu18;0.59413405 60 | 59;ip1;38.35533555 61 | 60;loss;0.3250066 62 | -------------------------------------------------------------------------------- /benchmark/Old Benchmarks/W9100_OPENCL/u_benchmark/train_forward_layers_timings.csv: -------------------------------------------------------------------------------- 1 | 0;label;0.05358175 2 | 1;data;0.0106764 3 | 2;silence;0.0062862 4 | 3;conv1;8.72105805 5 | 4;relu1;1.06838025 6 | 5;conv2;23.07337975 7 | 6;relu2;0.99218315 8 | 7;conv2_relu2_0_split;0.00231 9 | 8;pool1;1.30013475 10 | 9;conv3;10.09307355 11 | 10;relu3;0.49529725 12 | 11;conv4;17.069587 13 | 12;relu4;0.48062025 14 | 13;conv4_relu4_0_split;0.0044531 15 | 14;pool2;0.6306014 16 | 15;conv5;7.3282419 17 | 16;relu5;0.25926255 18 | 17;conv6;13.44869465 19 | 18;relu6;0.2546587 20 | 19;conv6_relu6_0_split;0.00442695 21 | 20;pool3;0.31381015 22 | 21;conv7;8.32209915 23 | 22;relu7;0.14536715 24 | 23;conv8;14.6917364 25 | 24;relu8;0.14136365 26 | 25;conv8_relu8_0_split;0.0042627 27 | 26;pool4;0.16154275 28 | 27;conv9;6.37048165 29 | 28;relu9;0.0886896 30 | 29;conv10;10.81720975 31 | 30;relu10;0.08405875 32 | 31;upconv1;14.9136653 33 | 32;conv11;2.76031885 34 | 33;mergecrop1;0.702784 35 | 34;conv12;26.1724679 36 | 35;relu11;0.12050855 37 | 36;conv13;12.23019055 38 | 37;relu12;0.1115402 39 | 38;upconv2;8.76163065 40 | 39;conv14;2.673114 41 | 40;mergecrop2;1.0999859 42 | 41;conv15;17.59355245 43 | 42;relu13;0.16732325 44 | 43;conv16;8.7310414 45 | 44;relu14;0.15539025 46 | 45;upconv3;6.9241866 47 | 46;conv17;2.52943365 48 | 47;mergecrop3;2.2199056 49 | 48;conv18;18.6912792 50 | 49;relu15;0.2590771 51 | 50;conv19;9.45790255 52 | 51;relu16;0.25249745 53 | 52;upconv4;9.92657805 54 | 53;conv20;2.4151179 55 | 54;mergecrop4;4.20427 56 | 55;conv21;20.4425451 57 | 56;relu17;0.4583843 58 | 57;conv22;10.5132404 59 | 58;relu18;0.4651783 60 | 59;ip1;1.15346165 61 | 60;loss;1.79207825 62 | -------------------------------------------------------------------------------- /benchmark/Old Benchmarks/W9100_OPENCL/u_benchmark/train_memory_usage.csv: -------------------------------------------------------------------------------- 1 | Peak memory usage;3115754464 2 | -------------------------------------------------------------------------------- /benchmark/Old Benchmarks/W9100_OPENCL/u_benchmark/train_total_timings.csv: -------------------------------------------------------------------------------- 1 | Forward;333.4063835 2 | Backward;756.5482909 3 | -------------------------------------------------------------------------------- /benchmark/Old Benchmarks/W9100_OPENCL/usk_benchmark/process_forward_layers_timings.csv: -------------------------------------------------------------------------------- 1 | 0;data;0.0272025 2 | 1;silence;0.0057302 3 | 2;conv1;18.1100799 4 | 3;relu1;2.10136265 5 | 4;conv2;43.28109185 6 | 5;relu2;1.55628205 7 | 6;conv2_relu2_0_split;0.0039665 8 | 7;pool1;2.02739385 9 | 8;conv3;15.39897875 10 | 9;relu3;0.76607075 11 | 10;conv4;26.26327995 12 | 11;relu4;0.7490438 13 | 12;conv4_relu4_0_split;0.00192105 14 | 13;pool2;0.97334865 15 | 14;conv5;19.000271 16 | 15;relu5;0.2902707 17 | 16;pool3;0.88451745 18 | 17;conv6;7.71049425 19 | 18;relu6;0.3061433 20 | 19;pool4;0.7766258 21 | 20;conv7;8.7051939 22 | 21;relu7;0.40265745 23 | 22;pool5;1.2807642 24 | 23;ip1;20.431142 25 | 24;relu8;0.5316599 26 | 25;ip2;3.9424277 27 | 26;relu9;0.25521325 28 | 27;upconv1;11.04719065 29 | 28;conv8;3.8784755 30 | 29;mergecrop1;3.8362237 31 | 30;conv9;28.14585245 32 | 31;relu10;0.42549475 33 | 32;conv10;14.98380395 34 | 33;relu11;0.41201255 35 | 34;upconv2;15.87372115 36 | 35;conv11;3.75866935 37 | 36;mergecrop2;7.34925065 38 | 37;conv12;34.5177475 39 | 38;relu12;0.80856805 40 | 39;conv13;21.69396605 41 | 40;relu13;0.8017454 42 | 41;ip3;1.9296221 43 | 42;prob;0.8962242 44 | -------------------------------------------------------------------------------- /benchmark/Old Benchmarks/W9100_OPENCL/usk_benchmark/process_memory_usage.csv: -------------------------------------------------------------------------------- 1 | Peak memory usage;5534664192 2 | -------------------------------------------------------------------------------- /benchmark/Old Benchmarks/W9100_OPENCL/usk_benchmark/process_total_timings.csv: -------------------------------------------------------------------------------- 1 | Forward;335.7636523 2 | -------------------------------------------------------------------------------- /benchmark/Old Benchmarks/W9100_OPENCL/usk_benchmark/train_backward_layers_timings.csv: -------------------------------------------------------------------------------- 1 | 0;label;0.020408 2 | 1;data;0.01954575 3 | 2;silence;0.01963255 4 | 3;conv1;33.99727525 5 | 4;relu1;1.9566463 6 | 5;conv2;147.7112473 7 | 6;relu2;1.9913576 8 | 7;conv2_relu2_0_split;1.9245861 9 | 8;pool1;8.0201947 10 | 9;conv3;49.9068698 11 | 10;relu3;0.93754255 12 | 11;conv4;56.26183905 13 | 12;relu4;1.08578375 14 | 13;conv4_relu4_0_split;1.03257075 15 | 14;pool2;3.9624403 16 | 15;conv5;57.55761535 17 | 16;relu5;0.3929708 18 | 17;pool3;1.2978679 19 | 18;conv6;59.6935311 20 | 19;relu6;0.37172355 21 | 20;pool4;1.21353995 22 | 21;conv7;95.7863797 23 | 22;relu7;0.51458385 24 | 23;pool5;1.9972711 25 | 24;ip1;276.1555573 26 | 25;relu8;0.5530381 27 | 26;ip2;33.49497165 28 | 27;relu9;0.31762845 29 | 28;upconv1;8.50593035 30 | 29;conv8;8.99847595 31 | 30;mergecrop1;2.13467515 32 | 31;conv9;69.3876522 33 | 32;relu10;0.55337425 34 | 33;conv10;67.26927205 35 | 34;relu11;0.59865495 36 | 35;upconv2;8.8253163 37 | 36;conv11;60.5732627 38 | 37;mergecrop2;4.06868735 39 | 38;conv12;166.0587509 40 | 39;relu12;1.04202455 41 | 40;conv13;187.7094839 42 | 41;relu13;1.0242664 43 | 42;ip3;105.5896153 44 | 43;loss;0.45912265 45 | -------------------------------------------------------------------------------- /benchmark/Old Benchmarks/W9100_OPENCL/usk_benchmark/train_forward_layers_timings.csv: -------------------------------------------------------------------------------- 1 | 0;label;0.02691855 2 | 1;data;0.0064373 3 | 2;silence;0.0030707 4 | 3;conv1;9.4694306 5 | 4;relu1;1.4191031 6 | 5;conv2;30.8471864 7 | 6;relu2;1.3981326 8 | 7;conv2_relu2_0_split;0.0036762 9 | 8;pool1;1.81421205 10 | 9;conv3;13.7496226 11 | 10;relu3;0.71205175 12 | 11;conv4;24.04618065 13 | 12;relu4;0.70497685 14 | 13;conv4_relu4_0_split;0.0021893 15 | 14;pool2;0.91543835 16 | 15;conv5;17.7445332 17 | 16;relu5;0.26786085 18 | 17;pool3;0.82905415 19 | 18;conv6;6.9440028 20 | 19;relu6;0.25664335 21 | 20;pool4;0.7235969 22 | 21;conv7;8.26808965 23 | 22;relu7;0.3723767 24 | 23;pool5;1.2396697 25 | 24;ip1;20.07890895 26 | 25;relu8;0.5030688 27 | 26;ip2;3.7313069 28 | 27;relu9;0.25017845 29 | 28;upconv1;10.97177575 30 | 29;conv8;3.7186867 31 | 30;mergecrop1;3.63074595 32 | 31;conv9;27.82885055 33 | 32;relu10;0.42961805 34 | 33;conv10;15.10763885 35 | 34;relu11;0.41886765 36 | 35;upconv2;15.21214715 37 | 36;conv11;3.8272214 38 | 37;mergecrop2;7.3590532 39 | 38;conv12;34.658337 40 | 39;relu12;0.81991195 41 | 40;conv13;21.64652125 42 | 41;relu13;0.79315655 43 | 42;ip3;2.0510735 44 | 43;loss;2.1361925 45 | -------------------------------------------------------------------------------- /benchmark/Old Benchmarks/W9100_OPENCL/usk_benchmark/train_memory_usage.csv: -------------------------------------------------------------------------------- 1 | Peak memory usage;5534664192 2 | -------------------------------------------------------------------------------- /benchmark/Old Benchmarks/W9100_OPENCL/usk_benchmark/train_total_timings.csv: -------------------------------------------------------------------------------- 1 | Forward;311.3539249 2 | Backward;1527.677162 3 | -------------------------------------------------------------------------------- /benchmark/Old Benchmarks/i7-4790K_OPENCL/sk_benchmark/process_forward_layers_timings.csv: -------------------------------------------------------------------------------- 1 | 0;data;0.0197348 2 | 1;silence;0.00346835 3 | 2;conv1;17.17629705 4 | 3;relu1;2.46946575 5 | 4;pool1;11.4977649 6 | 5;conv2;169.58097 7 | 6;relu2;4.56001075 8 | 7;pool2;35.22098255 9 | 8;conv3;169.8956321 10 | 9;relu3;5.82583305 11 | 10;pool3;47.7060709 12 | 11;ip1;3293.669467 13 | 12;relu4;13.55731985 14 | 13;ip2;144.3375851 15 | 14;relu5;5.7715019 16 | 15;ip3;64.47167365 17 | 16;prob;1.44359015 18 | -------------------------------------------------------------------------------- /benchmark/Old Benchmarks/i7-4790K_OPENCL/sk_benchmark/process_memory_usage.csv: -------------------------------------------------------------------------------- 1 | Peak memory usage;1672340488 2 | -------------------------------------------------------------------------------- /benchmark/Old Benchmarks/i7-4790K_OPENCL/sk_benchmark/process_total_timings.csv: -------------------------------------------------------------------------------- 1 | Forward;4011.37351 2 | -------------------------------------------------------------------------------- /benchmark/Old Benchmarks/i7-4790K_OPENCL/u_benchmark/process_forward_layers_timings.csv: -------------------------------------------------------------------------------- 1 | 0;data;0.0338416 2 | 1;silence;0.00376035 3 | 2;conv1;32.7535029 4 | 3;relu1;16.8183625 5 | 4;conv2;379.189203 6 | 5;relu2;14.0988721 7 | 6;conv2_relu2_0_split;0.00462115 8 | 7;pool1;40.56637945 9 | 8;conv3;108.2638719 10 | 9;relu3;6.2689461 11 | 10;conv4;252.4103339 12 | 11;relu4;6.2730168 13 | 12;conv4_relu4_0_split;0.0044319 14 | 13;pool2;14.2365974 15 | 14;conv5;84.2406006 16 | 15;relu5;2.7629116 17 | 16;conv6;149.3387747 18 | 17;relu6;3.7119462 19 | 18;conv6_relu6_0_split;0.0046957 20 | 19;pool3;5.9908442 21 | 20;conv7;41.16762615 22 | 21;relu7;1.21592 23 | 22;conv8;75.99790625 24 | 23;relu8;1.0760921 25 | 24;conv8_relu8_0_split;0.0047069 26 | 25;pool4;2.6488115 27 | 26;conv9;27.3658537 28 | 27;relu9;0.47050755 29 | 28;conv10;47.82995155 30 | 29;relu10;0.44853275 31 | 30;upconv1;35.4007902 32 | 31;conv11;10.49899 33 | 32;mergecrop1;21.4715123 34 | 33;conv12;110.9514104 35 | 34;relu11;0.6900561 36 | 35;conv13;45.68890595 37 | 36;relu12;0.614163 38 | 37;upconv2;61.1725004 39 | 38;conv14;11.9972047 40 | 39;mergecrop2;33.9450027 41 | 40;conv15;169.6663094 42 | 41;relu13;2.2528716 43 | 42;conv16;67.7111154 44 | 43;relu14;1.34378515 45 | 44;upconv3;92.25706355 46 | 45;conv17;17.0241159 47 | 46;mergecrop3;70.4215134 48 | 47;conv18;197.1291656 49 | 48;relu15;2.4057206 50 | 49;conv19;114.0431513 51 | 50;relu16;2.67084145 52 | 51;upconv4;151.9350133 53 | 52;conv20;19.70240335 54 | 53;mergecrop4;106.1428454 55 | 54;conv21;290.9787784 56 | 55;relu17;5.42952625 57 | 56;conv22;186.1338579 58 | 57;relu18;6.42344985 59 | 58;ip1;23.7616906 60 | 59;prob;10.0949773 61 | -------------------------------------------------------------------------------- /benchmark/Old Benchmarks/i7-4790K_OPENCL/u_benchmark/process_memory_usage.csv: -------------------------------------------------------------------------------- 1 | Peak memory usage;3382929120 2 | -------------------------------------------------------------------------------- /benchmark/Old Benchmarks/i7-4790K_OPENCL/u_benchmark/process_total_timings.csv: -------------------------------------------------------------------------------- 1 | Forward;3175.975714 2 | -------------------------------------------------------------------------------- /benchmark/Old Benchmarks/i7-4790K_OPENCL/u_benchmark/train_backward_layers_timings.csv: -------------------------------------------------------------------------------- 1 | 0;label;0.00181 2 | 1;data;0.001937 3 | 2;silence;0.0025248 4 | 3;conv1;70.32382755 5 | 4;relu1;13.68366265 6 | 5;conv2;680.7019759 7 | 6;relu2;13.3568728 8 | 7;conv2_relu2_0_split;17.3328649 9 | 8;pool1;132.6014684 10 | 9;conv3;228.5064239 11 | 10;relu3;6.4993343 12 | 11;conv4;411.8394184 13 | 12;relu4;6.37748825 14 | 13;conv4_relu4_0_split;8.5544971 15 | 14;pool2;80.8775803 16 | 15;conv5;130.1341153 17 | 16;relu5;6.00338535 18 | 17;conv6;287.1975124 19 | 18;relu6;5.7407019 20 | 19;conv6_relu6_0_split;8.6310786 21 | 20;pool3;50.26811215 22 | 21;conv7;79.79828075 23 | 22;relu7;2.38705955 24 | 23;conv8;164.3758726 25 | 24;relu8;2.86325875 26 | 25;conv8_relu8_0_split;3.11216685 27 | 26;pool4;18.4401272 28 | 27;conv9;58.42318965 29 | 28;relu9;0.8140326 30 | 29;conv10;105.5322685 31 | 30;relu10;0.6693809 32 | 31;upconv1;12.3612821 33 | 32;conv11;18.4842396 34 | 33;mergecrop1;20.62559535 35 | 34;conv12;227.7925385 36 | 35;relu11;1.56990655 37 | 36;conv13;101.4390026 38 | 37;relu12;1.3316387 39 | 38;upconv2;15.96770715 40 | 39;conv14;21.5701661 41 | 40;mergecrop2;21.92681435 42 | 41;conv15;304.3874991 43 | 42;relu13;3.30507105 44 | 43;conv16;111.6437974 45 | 44;relu14;2.0779874 46 | 45;upconv3;26.713927 47 | 46;conv17;25.79460575 48 | 47;mergecrop3;33.31448065 49 | 48;conv18;418.7071885 50 | 49;relu15;5.947056 51 | 50;conv19;192.4410715 52 | 51;relu16;4.121161 53 | 52;upconv4;53.54485185 54 | 53;conv20;35.64144925 55 | 54;mergecrop4;62.48256245 56 | 55;conv21;691.512851 57 | 56;relu17;6.63157035 58 | 57;conv22;342.3569081 59 | 58;relu18;8.7011129 60 | 59;ip1;18.82843765 61 | 60;loss;1.0856804 62 | -------------------------------------------------------------------------------- /benchmark/Old Benchmarks/i7-4790K_OPENCL/u_benchmark/train_forward_layers_timings.csv: -------------------------------------------------------------------------------- 1 | 0;label;0.026047 2 | 1;data;0.01465495 3 | 2;silence;0.003158 4 | 3;conv1;31.69898335 5 | 4;relu1;15.12982195 6 | 5;conv2;370.9914316 7 | 6;relu2;12.23004535 8 | 7;conv2_relu2_0_split;0.0044201 9 | 8;pool1;39.23394075 10 | 9;conv3;111.768493 11 | 10;relu3;5.46465335 12 | 11;conv4;261.7424803 13 | 12;relu4;7.2862877 14 | 13;conv4_relu4_0_split;0.00458585 15 | 14;pool2;15.7565823 16 | 15;conv5;83.635342 17 | 16;relu5;2.98370205 18 | 17;conv6;153.1930845 19 | 18;relu6;2.8637002 20 | 19;conv6_relu6_0_split;0.0045853 21 | 20;pool3;5.99291275 22 | 21;conv7;40.66043955 23 | 22;relu7;1.2249406 24 | 23;conv8;75.5369781 25 | 24;relu8;1.2255869 26 | 25;conv8_relu8_0_split;0.0044904 27 | 26;pool4;2.6753501 28 | 27;conv9;26.95606675 29 | 28;relu9;0.45483925 30 | 29;conv10;47.6312261 31 | 30;relu10;0.409097 32 | 31;upconv1;34.0826062 33 | 32;conv11;10.5685003 34 | 33;mergecrop1;21.69868405 35 | 34;conv12;111.3848496 36 | 35;relu11;0.75254985 37 | 36;conv13;45.826124 38 | 37;relu12;0.67891595 39 | 38;upconv2;60.4666953 40 | 39;conv14;11.72366045 41 | 40;mergecrop2;35.6398726 42 | 41;conv15;170.4436307 43 | 42;relu13;1.61695825 44 | 43;conv16;74.6783187 45 | 44;relu14;1.60882105 46 | 45;upconv3;96.7709341 47 | 46;conv17;16.89981545 48 | 47;mergecrop3;72.1473052 49 | 48;conv18;198.1538339 50 | 49;relu15;3.09899485 51 | 50;conv19;121.2900111 52 | 51;relu16;3.1666374 53 | 52;upconv4;155.1157924 54 | 53;conv20;19.9017498 55 | 54;mergecrop4;108.0388524 56 | 55;conv21;288.5121104 57 | 56;relu17;6.68511515 58 | 57;conv22;189.0321202 59 | 58;relu18;7.4455355 60 | 59;ip1;23.4751303 61 | 60;loss;12.4395392 62 | -------------------------------------------------------------------------------- /benchmark/Old Benchmarks/i7-4790K_OPENCL/u_benchmark/train_memory_usage.csv: -------------------------------------------------------------------------------- 1 | Peak memory usage;3382929120 2 | -------------------------------------------------------------------------------- /benchmark/Old Benchmarks/i7-4790K_OPENCL/u_benchmark/train_total_timings.csv: -------------------------------------------------------------------------------- 1 | Forward;3114.806026 2 | Backward;5159.989795 3 | -------------------------------------------------------------------------------- /benchmark/Old Benchmarks/i7-4790K_OPENCL/usk_benchmark/process_forward_layers_timings.csv: -------------------------------------------------------------------------------- 1 | 0;data;0.03487595 2 | 1;silence;0.00258945 3 | 2;conv1;48.61557245 4 | 3;relu1;32.7683139 5 | 4;conv2;648.0737625 6 | 5;relu2;26.12951175 7 | 6;conv2_relu2_0_split;0.0048252 8 | 7;pool1;55.1249391 9 | 8;conv3;153.5165482 10 | 9;relu3;10.656434 11 | 10;conv4;445.6779435 12 | 11;relu4;11.2459155 13 | 12;conv4_relu4_0_split;0.00491355 14 | 13;pool2;25.6329756 15 | 14;conv5;309.868844 16 | 15;relu5;2.56239655 17 | 16;pool3;20.9861795 18 | 17;conv6;100.2345962 19 | 18;relu6;1.87675845 20 | 19;pool4;15.3478669 21 | 20;conv7;119.2839036 22 | 21;relu7;3.4888729 23 | 22;pool5;28.252164 24 | 23;ip1;292.9182336 25 | 24;relu8;5.5181595 26 | 25;ip2;22.0443045 27 | 26;relu9;2.53075355 28 | 27;upconv1;141.599991 29 | 28;conv8;30.7605272 30 | 29;mergecrop1;95.82139615 31 | 30;conv9;405.6259046 32 | 31;relu10;3.94220585 33 | 32;conv10;248.3519226 34 | 33;relu11;3.8560956 35 | 34;upconv2;243.6505126 36 | 35;conv11;33.69031395 37 | 36;mergecrop2;162.277315 38 | 37;conv12;562.0382905 39 | 38;relu12;7.6691643 40 | 39;conv13;368.8477691 41 | 40;relu13;7.86040095 42 | 41;ip3;51.81376555 43 | 42;prob;14.7336286 44 | -------------------------------------------------------------------------------- /benchmark/Old Benchmarks/i7-4790K_OPENCL/usk_benchmark/process_memory_usage.csv: -------------------------------------------------------------------------------- 1 | Peak memory usage;5935565888 2 | -------------------------------------------------------------------------------- /benchmark/Old Benchmarks/i7-4790K_OPENCL/usk_benchmark/process_total_timings.csv: -------------------------------------------------------------------------------- 1 | Forward;4777.985872 2 | -------------------------------------------------------------------------------- /benchmark/Old Benchmarks/i7-4790K_OPENCL/usk_benchmark/train_backward_layers_timings.csv: -------------------------------------------------------------------------------- 1 | 0;label;0.00169915 2 | 1;data;0.00181785 3 | 2;silence;0.0023077 4 | 3;conv1;94.4982404 5 | 4;relu1;18.8624798 6 | 5;conv2;983.9052533 7 | 6;relu2;18.44694085 8 | 7;conv2_relu2_0_split;25.33598785 9 | 8;pool1;182.5268024 10 | 9;conv3;303.9893829 11 | 10;relu3;9.24266605 12 | 11;conv4;552.6715328 13 | 12;relu4;9.33223435 14 | 13;conv4_relu4_0_split;12.4099483 15 | 14;pool2;92.14017435 16 | 15;conv5;396.6936096 17 | 16;relu5;5.238711 18 | 17;pool3;26.93803495 19 | 18;conv6;122.4664839 20 | 19;relu6;5.3527607 21 | 20;pool4;25.4457205 22 | 21;conv7;157.5139276 23 | 22;relu7;3.81550255 24 | 23;pool5;26.82163455 25 | 24;ip1;348.743834 26 | 25;relu8;6.1089668 27 | 26;ip2;34.102442 28 | 27;relu9;3.0212909 29 | 28;upconv1;41.8773824 30 | 29;conv8;37.09202275 31 | 30;mergecrop1;55.7828302 32 | 31;conv9;643.9691761 33 | 32;relu10;5.48193835 34 | 33;conv10;354.1448383 35 | 34;relu11;5.55881085 36 | 35;upconv2;89.40449245 37 | 36;conv11;58.9613099 38 | 37;mergecrop2;108.0360175 39 | 38;conv12;1067.031948 40 | 39;relu12;10.6972071 41 | 40;conv13;683.758199 42 | 41;relu13;18.75664045 43 | 42;ip3;63.34360975 44 | 43;loss;5.76162865 45 | -------------------------------------------------------------------------------- /benchmark/Old Benchmarks/i7-4790K_OPENCL/usk_benchmark/train_forward_layers_timings.csv: -------------------------------------------------------------------------------- 1 | 0;label;0.03273945 2 | 1;data;0.0149574 3 | 2;silence;0.00335555 4 | 3;conv1;46.3682839 5 | 4;relu1;22.7003124 6 | 5;conv2;620.4459344 7 | 6;relu2;17.76886695 8 | 7;conv2_relu2_0_split;0.0044437 9 | 8;pool1;55.3475941 10 | 9;conv3;153.3155804 11 | 10;relu3;9.0759405 12 | 11;conv4;454.8608916 13 | 12;relu4;10.4069222 14 | 13;conv4_relu4_0_split;0.0043849 15 | 14;pool2;24.1447043 16 | 15;conv5;295.2389859 17 | 16;relu5;2.16665275 18 | 17;pool3;18.30254715 19 | 18;conv6;98.8176324 20 | 19;relu6;1.8880078 21 | 20;pool4;15.003768 22 | 21;conv7;120.6540203 23 | 22;relu7;3.35403295 24 | 23;pool5;29.8667171 25 | 24;ip1;298.2176655 26 | 25;relu8;5.7047535 27 | 26;ip2;22.1843636 28 | 27;relu9;3.27292625 29 | 28;upconv1;143.3331173 30 | 29;conv8;30.93230555 31 | 30;mergecrop1;96.59466075 32 | 31;conv9;405.1801721 33 | 32;relu10;4.8776539 34 | 33;conv10;251.9498786 35 | 34;relu11;4.66449775 36 | 35;upconv2;245.0451406 37 | 36;conv11;33.55757975 38 | 37;mergecrop2;162.296883 39 | 38;conv12;560.1788383 40 | 39;relu12;7.84960605 41 | 40;conv13;351.0315252 42 | 41;relu13;7.82098775 43 | 42;ip3;53.7303076 44 | 43;loss;17.14559895 45 | -------------------------------------------------------------------------------- /benchmark/Old Benchmarks/i7-4790K_OPENCL/usk_benchmark/train_memory_usage.csv: -------------------------------------------------------------------------------- 1 | Peak memory usage;5935565888 2 | -------------------------------------------------------------------------------- /benchmark/Old Benchmarks/i7-4790K_OPENCL/usk_benchmark/train_total_timings.csv: -------------------------------------------------------------------------------- 1 | Forward;4670.432751 2 | Backward;6781.203366 3 | -------------------------------------------------------------------------------- /benchmark/W9100_OPENCL/sk_benchmark/process_forward_layers_timings.csv: -------------------------------------------------------------------------------- 1 | 0;data;0.267114;0 2 | 1;silence;0.0169602;0 3 | 2;conv1;3.4850208;699388656 4 | 3;relu1;0.3369606;0 5 | 4;pool1;1.0029002;0 6 | 5;conv2;14.6427792;14062669312 7 | 6;relu2;0.4921696;0 8 | 7;pool2;1.557408;0 9 | 8;conv3;9.6329106;18401596416 10 | 9;relu3;0.5251826;0 11 | 10;pool3;1.961418;0 12 | 11;ip1;230.5576062;644228317184 13 | 12;relu4;0.9651192;0 14 | 13;ip2;10.0167458;17171480576 15 | 14;relu5;0.562539;0 16 | 15;ip3;1.5682372;33521664 17 | 16;prob;0.1528868;0 18 | -------------------------------------------------------------------------------- /benchmark/W9100_OPENCL/sk_benchmark/process_memory_usage.csv: -------------------------------------------------------------------------------- 1 | Peak memory usage;2892571592 2 | data;629292 3 | label;4 4 | conv1;9547968 5 | pool1;9462528 6 | conv2;23447552 7 | pool2;23011328 8 | conv3;31961088 9 | pool3;30720000 10 | ip1;67108864 11 | ip2;33554432 12 | ip3;131072 13 | prob;131072 14 | -------------------------------------------------------------------------------- /benchmark/W9100_OPENCL/sk_benchmark/process_total_timings.csv: -------------------------------------------------------------------------------- 1 | Forward;275.3027542 2 | -------------------------------------------------------------------------------- /benchmark/W9100_OPENCL/sk_benchmark/train_backward_layers_timings.csv: -------------------------------------------------------------------------------- 1 | 0;label;0.0597106;0 2 | 1;data;0.064706;0 3 | 2;silence;0.068661;0 4 | 3;conv1;18.407904;0 5 | 4;relu1;0.336475;0 6 | 5;pool1;0.9681864;0 7 | 6;conv2;36.5485722;0 8 | 7;relu2;0.545752;0 9 | 8;pool2;2.1492892;0 10 | 9;conv3;23.6169218;0 11 | 10;relu3;0.7426274;0 12 | 11;pool3;3.1105394;0 13 | 12;ip1;511.0854038;0 14 | 13;relu4;1.1232222;0 15 | 14;ip2;36.9590878;0 16 | 15;relu5;0.5654822;0 17 | 16;ip3;11.9356038;0 18 | 17;loss;0.1615142;0 19 | -------------------------------------------------------------------------------- /benchmark/W9100_OPENCL/sk_benchmark/train_forward_layers_timings.csv: -------------------------------------------------------------------------------- 1 | 0;label;0.31516;0 2 | 1;data;0.009948;0 3 | 2;silence;0.0045286;0 4 | 3;conv1;3.3664376;699388656 5 | 4;relu1;0.3310712;0 6 | 5;pool1;0.99215;0 7 | 6;conv2;13.6490884;14062669312 8 | 7;relu2;0.4264744;0 9 | 8;pool2;1.5393352;0 10 | 9;conv3;9.5149946;18401596416 11 | 10;relu3;0.5326766;0 12 | 11;pool3;1.910301;0 13 | 12;ip1;229.7776796;644228317184 14 | 13;relu4;0.974902;0 15 | 14;ip2;9.9059656;17171480576 16 | 15;relu5;0.5180868;0 17 | 16;ip3;1.615396;33521664 18 | 17;loss;0.729353;0 19 | -------------------------------------------------------------------------------- /benchmark/W9100_OPENCL/sk_benchmark/train_memory_usage.csv: -------------------------------------------------------------------------------- 1 | Peak memory usage;3204621764 2 | label;65536 3 | labeli;4 4 | data;629292 5 | datai;4 6 | conv1;9547968 7 | pool1;9462528 8 | conv2;23447552 9 | pool2;23011328 10 | conv3;31961088 11 | pool3;30720000 12 | ip1;67108864 13 | ip2;33554432 14 | ip3;131072 15 | (automatic);4 16 | -------------------------------------------------------------------------------- /benchmark/W9100_OPENCL/sk_benchmark/train_total_timings.csv: -------------------------------------------------------------------------------- 1 | Forward;276.4525018 2 | Backward;650.2205298 3 | -------------------------------------------------------------------------------- /benchmark/W9100_OPENCL/u_benchmark/process_forward_layers_timings.csv: -------------------------------------------------------------------------------- 1 | 0;data;0.0384086;0 2 | 1;silence;0.0211522;0 3 | 2;conv1;7.2544112;1102060800 4 | 3;relu1;0.9899186;0 5 | 4;conv2;21.6811124;23765774336 6 | 5;relu2;1.0370588;0 7 | 6;conv2_relu2_0_split;0.0316092;0 8 | 7;pool1;1.315202;0 9 | 8;conv3;10.0028218;11716111872 10 | 9;relu3;0.512285;0 11 | 10;conv4;16.4709;23111065600 12 | 11;relu4;0.532091;0 13 | 12;conv4_relu4_0_split;0.0217948;0 14 | 13;pool2;0.6792902;0 15 | 14;conv5;7.5692236;11227732992 16 | 15;relu5;0.2856308;0 17 | 16;conv6;12.9858948;21814034432 18 | 17;relu6;0.2620402;0 19 | 18;conv6_relu6_0_split;0.0113666;0 20 | 19;pool3;0.3233584;0 21 | 20;conv7;8.0643454;10274863104 22 | 21;relu7;0.189861;0 23 | 22;conv8;14.2658316;19325255680 24 | 23;relu8;0.1650048;0 25 | 24;conv8_relu8_0_split;0.0117044;0 26 | 25;pool4;0.1797362;0 27 | 26;conv9;6.9132262;8492544000 28 | 27;relu9;0.1324578;0 29 | 28;conv10;11.4678026;14796701696 30 | 29;relu10;0.126174;0 31 | 30;upconv1;16.2120674;0 32 | 31;conv11;3.6529848;3286728704 33 | 32;mergecrop1;0.769704;0 34 | 33;conv12;25.8365688;27517335552 35 | 34;relu11;0.1550806;0 36 | 35;conv13;12.098201;12757688320 37 | 36;relu12;0.1611028;0 38 | 37;upconv2;9.44749;0 39 | 38;conv14;3.0775862;2832580608 40 | 39;mergecrop2;1.136031;0 41 | 40;conv15;17.0083568;24543452160 42 | 41;relu13;0.187518;0 43 | 42;conv16;8.389239;11793920000 44 | 43;relu14;0.178848;0 45 | 44;upconv3;7.325522;0 46 | 45;conv17;2.7302628;2616320000 47 | 46;mergecrop3;2.2043876;0 48 | 47;conv18;18.1398904;23118441984 49 | 48;relu15;0.2931772;0 50 | 49;conv19;9.224855;11324422144 51 | 50;relu16;0.2911886;0 52 | 51;upconv4;9.781852;0 53 | 52;conv20;2.8940774;2507796480 54 | 53;mergecrop4;4.0675874;0 55 | 54;conv21;20.7526626;22418323200 56 | 55;relu17;0.5069742;0 57 | 56;conv22;10.8709292;11089673216 58 | 57;relu18;0.488177;0 59 | 58;ip1;0.9263292;38238176 60 | 59;prob;0.1962042;0 61 | -------------------------------------------------------------------------------- /benchmark/W9100_OPENCL/u_benchmark/process_memory_usage.csv: -------------------------------------------------------------------------------- 1 | Peak memory usage;1793039512 2 | data;3926208 3 | datai;4 4 | conv1;83174400 5 | conv2;82591744 6 | conv2_relu2_0_split_0;82591744 7 | conv2_relu2_0_split_1;82591744 8 | pool1;20647936 9 | conv3;40716288 10 | conv4;40140800 11 | conv4_relu4_0_split_0;40140800 12 | conv4_relu4_0_split_1;40140800 13 | pool2;10035200 14 | conv5;19501056 15 | conv6;18939904 16 | conv6_relu6_0_split_0;18939904 17 | conv6_relu6_0_split_1;18939904 18 | pool3;4734976 19 | conv7;8921088 20 | conv8;8388608 21 | conv8_relu8_0_split_0;8388608 22 | conv8_relu8_0_split_1;8388608 23 | pool4;2097152 24 | conv9;3686400 25 | conv10;3211264 26 | upconv1;12845056 27 | conv11;6422528 28 | mergecrop1;12845056 29 | conv12;5971968 30 | conv13;5537792 31 | upconv2;22151168 32 | conv14;11075584 33 | mergecrop2;22151168 34 | conv15;10653696 35 | conv16;10240000 36 | upconv3;40960000 37 | conv17;20480000 38 | mergecrop3;40960000 39 | conv18;20072448 40 | conv19;19668992 41 | upconv4;78675968 42 | conv20;39337984 43 | mergecrop4;78675968 44 | conv21;38937600 45 | conv22;38539264 46 | ip1;1204352 47 | prob;1204352 48 | -------------------------------------------------------------------------------- /benchmark/W9100_OPENCL/u_benchmark/process_total_timings.csv: -------------------------------------------------------------------------------- 1 | Forward;311.1919932 2 | -------------------------------------------------------------------------------- /benchmark/W9100_OPENCL/u_benchmark/train_backward_layers_timings.csv: -------------------------------------------------------------------------------- 1 | 0;label;0.003739;0 2 | 1;data;0.0039258;0 3 | 2;silence;0.006739;0 4 | 3;conv1;22.7903924;0 5 | 4;relu1;1.282299;0 6 | 5;conv2;92.4920406;0 7 | 6;relu2;1.4300242;0 8 | 7;conv2_relu2_0_split;1.3799772;0 9 | 8;pool1;5.530885;0 10 | 9;conv3;33.1254904;0 11 | 10;relu3;0.6586884;0 12 | 11;conv4;36.3971962;0 13 | 12;relu4;0.7372244;0 14 | 13;conv4_relu4_0_split;0.702916;0 15 | 14;pool2;2.712687;0 16 | 15;conv5;20.123048;0 17 | 16;relu5;0.3644274;0 18 | 17;conv6;19.7487388;0 19 | 18;relu6;0.3707866;0 20 | 19;conv6_relu6_0_split;0.3134476;0 21 | 20;pool3;1.2977584;0 22 | 21;conv7;13.6065934;0 23 | 22;relu7;0.1779002;0 24 | 23;conv8;20.5961;0 25 | 24;relu8;0.1829554;0 26 | 25;conv8_relu8_0_split;0.163696;0 27 | 26;pool4;0.6213156;0 28 | 27;conv9;11.1597182;0 29 | 28;relu9;0.147524;0 30 | 29;conv10;10.2178382;0 31 | 30;relu10;0.1734522;0 32 | 31;upconv1;17.1727114;0 33 | 32;conv11;3.5229796;0 34 | 33;mergecrop1;0.6684512;0 35 | 34;conv12;33.7325672;0 36 | 35;relu11;0.1623016;0 37 | 36;conv13;10.5613716;0 38 | 37;relu12;0.2142108;0 39 | 38;upconv2;9.2769812;0 40 | 39;conv14;4.3359708;0 41 | 40;mergecrop2;1.1593732;0 42 | 41;conv15;42.3912136;0 43 | 42;relu13;0.2308758;0 44 | 43;conv16;10.9977612;0 45 | 44;relu14;0.259242;0 46 | 45;upconv3;6.3393952;0 47 | 46;conv17;10.0888704;0 48 | 47;mergecrop3;1.9856034;0 49 | 48;conv18;66.7833362;0 50 | 49;relu15;0.3717014;0 51 | 50;conv19;17.5924308;0 52 | 51;relu16;0.4273248;0 53 | 52;upconv4;6.8450436;0 54 | 53;conv20;32.3812722;0 55 | 54;mergecrop4;3.6278156;0 56 | 55;conv21;86.340569;0 57 | 56;relu17;0.6978062;0 58 | 57;conv22;43.7455824;0 59 | 58;relu18;0.6457996;0 60 | 59;ip1;36.7875752;0 61 | 60;loss;0.5952816;0 62 | -------------------------------------------------------------------------------- /benchmark/W9100_OPENCL/u_benchmark/train_forward_layers_timings.csv: -------------------------------------------------------------------------------- 1 | 0;label;0.1164118;0 2 | 1;data;0.0086184;0 3 | 2;silence;0.0035884;0 4 | 3;conv1;11.0395416;1102060800 5 | 4;relu1;1.4056194;0 6 | 5;conv2;26.3168936;23765774336 7 | 6;relu2;1.1574734;0 8 | 7;conv2_relu2_0_split;0.063485;0 9 | 8;pool1;1.4243122;0 10 | 9;conv3;10.311091;11716111872 11 | 10;relu3;0.5993506;0 12 | 11;conv4;16.4911678;23111065600 13 | 12;relu4;0.523825;0 14 | 13;conv4_relu4_0_split;0.017851;0 15 | 14;pool2;0.6486854;0 16 | 15;conv5;6.9475458;11227732992 17 | 16;relu5;0.265699;0 18 | 17;conv6;12.7320132;21814034432 19 | 18;relu6;0.3144884;0 20 | 19;conv6_relu6_0_split;0.0372782;0 21 | 20;pool3;0.3426722;0 22 | 21;conv7;7.8604838;10274863104 23 | 22;relu7;0.1630536;0 24 | 23;conv8;13.9521812;19325255680 25 | 24;relu8;0.1634642;0 26 | 25;conv8_relu8_0_split;0.0122554;0 27 | 26;pool4;0.1717706;0 28 | 27;conv9;6.7290878;8492544000 29 | 28;relu9;0.1088368;0 30 | 29;conv10;11.354158;14796701696 31 | 30;relu10;0.1023858;0 32 | 31;upconv1;16.100311;0 33 | 32;conv11;3.3501648;3286728704 34 | 33;mergecrop1;0.7160038;0 35 | 34;conv12;25.875685;27517335552 36 | 35;relu11;0.160474;0 37 | 36;conv13;11.9677378;12757688320 38 | 37;relu12;0.1524584;0 39 | 38;upconv2;9.3753654;0 40 | 39;conv14;2.8116018;2832580608 41 | 40;mergecrop2;1.170809;0 42 | 41;conv15;16.859956;24543452160 43 | 42;relu13;0.1910946;0 44 | 43;conv16;8.432135;11793920000 45 | 44;relu14;0.1996394;0 46 | 45;upconv3;7.3742686;0 47 | 46;conv17;2.6933364;2616320000 48 | 47;mergecrop3;2.1792764;0 49 | 48;conv18;17.7995038;23118441984 50 | 49;relu15;0.3177418;0 51 | 50;conv19;9.1808268;11324422144 52 | 51;relu16;0.2993054;0 53 | 52;upconv4;10.308265;0 54 | 53;conv20;2.7332296;2507796480 55 | 54;mergecrop4;4.6643916;0 56 | 55;conv21;20.5199534;22418323200 57 | 56;relu17;0.4766196;0 58 | 57;conv22;10.5977968;11089673216 59 | 58;relu18;0.4634516;0 60 | 59;ip1;0.9123474;38238176 61 | 60;loss;1.0046452;0 62 | -------------------------------------------------------------------------------- /benchmark/W9100_OPENCL/u_benchmark/train_memory_usage.csv: -------------------------------------------------------------------------------- 1 | Peak memory usage;3098889896 2 | label;602176 3 | labeli;4 4 | data;3926208 5 | datai;4 6 | conv1;83174400 7 | conv2;82591744 8 | conv2_relu2_0_split_0;82591744 9 | conv2_relu2_0_split_1;82591744 10 | pool1;20647936 11 | conv3;40716288 12 | conv4;40140800 13 | conv4_relu4_0_split_0;40140800 14 | conv4_relu4_0_split_1;40140800 15 | pool2;10035200 16 | conv5;19501056 17 | conv6;18939904 18 | conv6_relu6_0_split_0;18939904 19 | conv6_relu6_0_split_1;18939904 20 | pool3;4734976 21 | conv7;8921088 22 | conv8;8388608 23 | conv8_relu8_0_split_0;8388608 24 | conv8_relu8_0_split_1;8388608 25 | pool4;2097152 26 | conv9;3686400 27 | conv10;3211264 28 | upconv1;12845056 29 | conv11;6422528 30 | mergecrop1;12845056 31 | conv12;5971968 32 | conv13;5537792 33 | upconv2;22151168 34 | conv14;11075584 35 | mergecrop2;22151168 36 | conv15;10653696 37 | conv16;10240000 38 | upconv3;40960000 39 | conv17;20480000 40 | mergecrop3;40960000 41 | conv18;20072448 42 | conv19;19668992 43 | upconv4;78675968 44 | conv20;39337984 45 | mergecrop4;78675968 46 | conv21;38937600 47 | conv22;38539264 48 | ip1;1204352 49 | (automatic);4 50 | -------------------------------------------------------------------------------- /benchmark/W9100_OPENCL/u_benchmark/train_total_timings.csv: -------------------------------------------------------------------------------- 1 | Forward;319.3320198 2 | Backward;746.6637294 3 | -------------------------------------------------------------------------------- /benchmark/W9100_OPENCL/usk_benchmark/process_forward_layers_timings.csv: -------------------------------------------------------------------------------- 1 | 0;data;0.02421;0 2 | 1;silence;0.0046744;0 3 | 2;conv1;4.866212;638903552 4 | 3;relu1;0.6315538;0 5 | 4;conv2;12.1766318;13747470336 6 | 5;relu2;0.5883846;0 7 | 6;conv2_relu2_0_split;0.0102962;0 8 | 7;pool1;0.7501166;0 9 | 8;conv3;18.5130444;26253855616 10 | 9;relu3;0.3806416;0 11 | 10;pool2;1.3158998;0 12 | 11;conv4;15.4242522;21813442560 13 | 12;relu4;0.3886042;0 14 | 13;pool3;1.2142736;0 15 | 14;conv5;14.4862138;18922176000 16 | 15;relu5;0.3857272;0 17 | 16;pool4;1.0849694;0 18 | 17;ip1;58.8956638;141758822400 19 | 18;relu6;0.5437024;0 20 | 19;ip2;3.5903138;4425907200 21 | 20;relu7;0.2550674;0 22 | 21;upconv1;11.0575026;0 23 | 22;conv6;3.5811492;4421580800 24 | 23;mergecrop1;3.2649672;0 25 | 24;conv7;20.5726518;29437263360 26 | 25;relu8;0.4243262;0 27 | 26;conv8;9.9952966;9659482112 28 | 27;relu9;0.2309326;0 29 | 28;ip3;0.5980138;16646144 30 | 29;prob;0.13386;0 31 | -------------------------------------------------------------------------------- /benchmark/W9100_OPENCL/usk_benchmark/process_memory_usage.csv: -------------------------------------------------------------------------------- 1 | Peak memory usage;1699750972 2 | data;2281152 3 | datai;4 4 | conv1;48219136 5 | conv2;47775744 6 | conv2_relu2_0_split_0;47775744 7 | conv2_relu2_0_split_1;47775744 8 | pool1;11943936 9 | conv3;22794752 10 | pool2;22579200 11 | conv4;21307392 12 | pool3;20891648 13 | conv5;18483200 14 | pool4;17713152 15 | ip1;34611200 16 | ip2;17305600 17 | upconv1;69222400 18 | conv6;34611200 19 | mergecrop1;51916800 20 | conv7;34080768 21 | conv8;16777216 22 | ip3;524288 23 | prob;524288 24 | -------------------------------------------------------------------------------- /benchmark/W9100_OPENCL/usk_benchmark/process_total_timings.csv: -------------------------------------------------------------------------------- 1 | Forward;184.6787066 2 | -------------------------------------------------------------------------------- /benchmark/W9100_OPENCL/usk_benchmark/train_backward_layers_timings.csv: -------------------------------------------------------------------------------- 1 | 0;label;0.009055;0 2 | 1;data;0.0225334;0 3 | 2;silence;0.0196054;0 4 | 3;conv1;13.8204502;0 5 | 4;relu1;0.8234976;0 6 | 5;conv2;54.1101926;0 7 | 6;relu2;0.900711;0 8 | 7;conv2_relu2_0_split;0.8464584;0 9 | 8;pool1;3.307561;0 10 | 9;conv3;189.2290574;0 11 | 10;relu3;0.4696566;0 12 | 11;pool2;2.0700584;0 13 | 12;conv4;33.3275978;0 14 | 13;relu4;0.4578572;0 15 | 14;pool3;1.9540708;0 16 | 15;conv5;55.3389328;0 17 | 16;relu5;0.429952;0 18 | 17;pool4;1.6791424;0 19 | 18;ip1;473.054469;0 20 | 19;relu6;0.5511518;0 21 | 20;ip2;8.698279;0 22 | 21;relu7;0.3181208;0 23 | 22;upconv1;7.9862598;0 24 | 23;conv6;15.9697512;0 25 | 24;mergecrop1;2.4973894;0 26 | 25;conv7;103.877423;0 27 | 26;relu8;0.589231;0 28 | 27;conv8;49.7956402;0 29 | 28;relu9;0.3686352;0 30 | 29;ip3;18.387722;0 31 | 30;loss;0.387076;0 32 | -------------------------------------------------------------------------------- /benchmark/W9100_OPENCL/usk_benchmark/train_forward_layers_timings.csv: -------------------------------------------------------------------------------- 1 | 0;label;0.0210318;0 2 | 1;data;0.007176;0 3 | 2;silence;0.0028636;0 4 | 3;conv1;5.3016472;638903552 5 | 4;relu1;0.679872;0 6 | 5;conv2;12.3801446;13747470336 7 | 6;relu2;0.6189914;0 8 | 7;conv2_relu2_0_split;0.0109976;0 9 | 8;pool1;0.7870858;0 10 | 9;conv3;18.5637834;26253855616 11 | 10;relu3;0.3842648;0 12 | 11;pool2;1.297485;0 13 | 12;conv4;15.3843472;21813442560 14 | 13;relu4;0.3601628;0 15 | 14;pool3;1.2250246;0 16 | 15;conv5;14.4909344;18922176000 17 | 16;relu5;0.353725;0 18 | 17;pool4;1.062192;0 19 | 18;ip1;58.888826;141758822400 20 | 19;relu6;0.5205064;0 21 | 20;ip2;3.4008608;4425907200 22 | 21;relu7;0.2703278;0 23 | 22;upconv1;10.5978498;0 24 | 23;conv6;3.459726;4421580800 25 | 24;mergecrop1;3.0804746;0 26 | 25;conv7;20.7551948;29437263360 27 | 26;relu8;0.4450632;0 28 | 27;conv8;10.1331528;9659482112 29 | 28;relu9;0.255616;0 30 | 29;ip3;0.7966742;16646144 31 | 30;loss;0.7015518;0 32 | -------------------------------------------------------------------------------- /benchmark/W9100_OPENCL/usk_benchmark/train_memory_usage.csv: -------------------------------------------------------------------------------- 1 | Peak memory usage;2311182092 2 | label;262144 3 | labeli;4 4 | data;2281152 5 | datai;4 6 | conv1;48219136 7 | conv2;47775744 8 | conv2_relu2_0_split_0;47775744 9 | conv2_relu2_0_split_1;47775744 10 | pool1;11943936 11 | conv3;22794752 12 | pool2;22579200 13 | conv4;21307392 14 | pool3;20891648 15 | conv5;18483200 16 | pool4;17713152 17 | ip1;34611200 18 | ip2;17305600 19 | upconv1;69222400 20 | conv6;34611200 21 | mergecrop1;51916800 22 | conv7;34080768 23 | conv8;16777216 24 | ip3;524288 25 | (automatic);4 26 | -------------------------------------------------------------------------------- /benchmark/W9100_OPENCL/usk_benchmark/train_total_timings.csv: -------------------------------------------------------------------------------- 1 | Forward;185.583737 2 | Backward;1048.855723 3 | -------------------------------------------------------------------------------- /benchmark/W9100_OPENCL/usk_benchmark_512/process_forward_layers_timings.csv: -------------------------------------------------------------------------------- 1 | 0;data;0.04917;0 2 | 1;silence;0.0207078;0 3 | 2;conv1;12.0855516;1614931200 4 | 3;relu1;1.7676192;0 5 | 4;conv2;34.8133522;34868412416 6 | 5;relu2;1.4742244;0 7 | 6;conv2_relu2_0_split;0.0339204;0 8 | 7;pool1;1.9107806;0 9 | 8;conv3;41.7153594;67768454016 10 | 9;relu3;0.8861996;0 11 | 10;pool2;3.2547032;0 12 | 11;conv4;36.2176486;57775011840 13 | 12;relu4;0.8619388;0 14 | 13;pool3;3.0945682;0 15 | 14;conv5;34.2234438;53005155840 16 | 15;relu5;0.8061624;0 17 | 16;pool4;2.8793208;0 18 | 17;ip1;234.9312972;558345222144 19 | 18;relu6;1.7617316;0 20 | 19;ip2;8.9878584;17432312832 21 | 20;relu7;0.9185526;0 22 | 21;upconv1;30.9896536;0 23 | 22;conv6;11.1260828;17415272448 24 | 23;mergecrop1;11.2061804;0 25 | 24;conv7;73.6487964;116838039040 26 | 25;relu8;1.7160592;0 27 | 26;conv8;41.8129304;38637928448 28 | 27;relu9;0.8734104;0 29 | 28;ip3;2.726206;66584576 30 | 29;prob;0.2934764;0 31 | -------------------------------------------------------------------------------- /benchmark/W9100_OPENCL/usk_benchmark_512/process_memory_usage.csv: -------------------------------------------------------------------------------- 1 | Peak memory usage;6225709116 2 | data;5746368 3 | datai;4 4 | conv1;121881600 5 | conv2;121176064 6 | conv2_relu2_0_split_0;121176064 7 | conv2_relu2_0_split_1;121176064 8 | pool1;30294016 9 | conv3;58839552 10 | pool2;58492928 11 | conv4;56434688 12 | pool3;55756800 13 | conv5;51775488 14 | pool4;50481152 15 | ip1;136323072 16 | ip2;68161536 17 | upconv1;272646144 18 | conv6;136323072 19 | mergecrop1;204484608 20 | conv7;135268352 21 | conv8;67108864 22 | ip3;2097152 23 | prob;2097152 24 | -------------------------------------------------------------------------------- /benchmark/W9100_OPENCL/usk_benchmark_512/process_total_timings.csv: -------------------------------------------------------------------------------- 1 | Forward;595.9614998 2 | -------------------------------------------------------------------------------- /benchmark/W9100_OPENCL/usk_benchmark_512/train_backward_layers_timings.csv: -------------------------------------------------------------------------------- 1 | 0;label;0.0554826;0 2 | 1;data;0.0567094;0 3 | 2;silence;0.074131;0 4 | 3;conv1;34.0451052;0 5 | 4;relu1;1.8940132;0 6 | 5;conv2;135.288941;0 7 | 6;relu2;2.0505562;0 8 | 7;conv2_relu2_0_split;2.0056512;0 9 | 8;pool1;8.1398132;0 10 | 9;conv3;488.543611;0 11 | 10;relu3;1.1051124;0 12 | 11;pool2;5.0387972;0 13 | 12;conv4;94.8176062;0 14 | 13;relu4;1.0652066;0 15 | 14;pool3;4.8958614;0 16 | 15;conv5;177.4987414;0 17 | 16;relu5;1.0093002;0 18 | 17;pool4;4.631934;0 19 | 18;ip1;2752.894852;0 20 | 19;relu6;2.0450826;0 21 | 20;ip2;47.830296;0 22 | 21;relu7;1.1806924;0 23 | 22;upconv1;18.9867796;0 24 | 23;conv6;59.7599066;0 25 | 24;mergecrop1;9.4122604;0 26 | 25;conv7;437.1023758;0 27 | 26;relu8;2.04888;0 28 | 27;conv8;233.8411938;0 29 | 28;relu9;1.1549644;0 30 | 29;ip3;98.5092862;0 31 | 30;loss;0.6675912;0 32 | -------------------------------------------------------------------------------- /benchmark/W9100_OPENCL/usk_benchmark_512/train_forward_layers_timings.csv: -------------------------------------------------------------------------------- 1 | 0;label;0.023046;0 2 | 1;data;0.0081492;0 3 | 2;silence;0.0027256;0 4 | 3;conv1;9.264682;1614931200 5 | 4;relu1;1.4491014;0 6 | 5;conv2;31.0880892;34868412416 7 | 6;relu2;1.4713426;0 8 | 7;conv2_relu2_0_split;0.0319252;0 9 | 8;pool1;1.8498008;0 10 | 9;conv3;41.6876142;67768454016 11 | 10;relu3;0.8495948;0 12 | 11;pool2;3.215349;0 13 | 12;conv4;36.2822714;57775011840 14 | 13;relu4;0.8261364;0 15 | 14;pool3;3.0824942;0 16 | 15;conv5;34.6468132;53005155840 17 | 16;relu5;0.7442834;0 18 | 17;pool4;2.8088508;0 19 | 18;ip1;227.1143374;558345222144 20 | 19;relu6;1.6821952;0 21 | 20;ip2;8.5901944;17432312832 22 | 21;relu7;0.872874;0 23 | 22;upconv1;30.9192762;0 24 | 23;conv6;11.0372804;17415272448 25 | 24;mergecrop1;9.7548626;0 26 | 25;conv7;72.9995972;116838039040 27 | 26;relu8;1.6517988;0 28 | 27;conv8;41.0164854;38637928448 29 | 28;relu9;0.8668784;0 30 | 29;ip3;2.1371512;66584576 31 | 30;loss;1.429576;0 32 | -------------------------------------------------------------------------------- /benchmark/W9100_OPENCL/usk_benchmark_512/train_memory_usage.csv: -------------------------------------------------------------------------------- 1 | Peak memory usage;8126552844 2 | label;1048576 3 | labeli;4 4 | data;5746368 5 | datai;4 6 | conv1;121881600 7 | conv2;121176064 8 | conv2_relu2_0_split_0;121176064 9 | conv2_relu2_0_split_1;121176064 10 | pool1;30294016 11 | conv3;58839552 12 | pool2;58492928 13 | conv4;56434688 14 | pool3;55756800 15 | conv5;51775488 16 | pool4;50481152 17 | ip1;136323072 18 | ip2;68161536 19 | upconv1;272646144 20 | conv6;136323072 21 | mergecrop1;204484608 22 | conv7;135268352 23 | conv8;67108864 24 | ip3;2097152 25 | (automatic);4 26 | -------------------------------------------------------------------------------- /benchmark/W9100_OPENCL/usk_benchmark_512/train_total_timings.csv: -------------------------------------------------------------------------------- 1 | Forward;588.0471572 2 | Backward;4626.000461 3 | -------------------------------------------------------------------------------- /benchmark/benchmark_sk.sh: -------------------------------------------------------------------------------- 1 | ./../../caffe_neural_tool/build/caffe_neural_tool --gpu 2 --benchmark 0 --proto 'train_process_sk_2.prototxt' 2 | -------------------------------------------------------------------------------- /benchmark/benchmark_u.sh: -------------------------------------------------------------------------------- 1 | ./../../caffe_neural_tool/build/caffe_neural_tool --gpu 0 --benchmark 0 --proto 'train_process_u_2.prototxt' 2 | -------------------------------------------------------------------------------- /benchmark/benchmark_u_small.sh: -------------------------------------------------------------------------------- 1 | ./../../caffe_neural_tool/build/caffe_neural_tool --gpu 0 --benchmark 0 --proto 'train_process_u_2small.prototxt' 2 | -------------------------------------------------------------------------------- /benchmark/benchmark_usk.sh: -------------------------------------------------------------------------------- 1 | ./../../caffe_neural_tool/build/caffe_neural_tool --gpu 3 --benchmark 0 --proto 'train_process_usk_2.prototxt' 2 | -------------------------------------------------------------------------------- /benchmark/clBLAS-BENCH/bench.sh: -------------------------------------------------------------------------------- 1 | #!/bin/bash 2 | for ((i=6; i<13; i+=1)); do 3 | let mnk=(2**$i) 4 | echo "BLAS-client -o1 --transposeB 1 -m $mnk -n $mnk -k $mnk" >> bench.txt 5 | clBLAS-client -o1 --transposeB 1 -m $mnk -n $mnk -k $mnk >> bench.txt 6 | done 7 | 8 | for ((i=1; i<13; i+=1)); do 9 | let mnk=(512*$i) 10 | echo "BLAS-client -o1 --transposeB 1 -m $mnk -n $mnk -k $mnk" >> bench.txt 11 | clBLAS-client -o1 --transposeB 1 -m $mnk -n $mnk -k $mnk >> bench.txt 12 | done 13 | 14 | -------------------------------------------------------------------------------- /benchmark/clBLAS-BENCH/bench.txt.old: -------------------------------------------------------------------------------- 1 | BLAS-client -o1 --transposeB 1 -m 64 -n 64 -k 64 2 | BLAS kernel execution time < ns >: 30425 3 | BLAS kernel execution Gflops < 2.0*M*N*K/time >: 17.2321 4 | BLAS-client -o1 --transposeB 1 -m 128 -n 128 -k 128 5 | BLAS kernel execution time < ns >: 36363.2 6 | BLAS kernel execution Gflops < 2.0*M*N*K/time >: 115.345 7 | BLAS-client -o1 --transposeB 1 -m 256 -n 256 -k 256 8 | BLAS kernel execution time < ns >: 54394.7 9 | BLAS kernel execution Gflops < 2.0*M*N*K/time >: 616.869 10 | BLAS-client -o1 --transposeB 1 -m 512 -n 512 -k 512 11 | BLAS kernel execution time < ns >: 207337 12 | BLAS kernel execution Gflops < 2.0*M*N*K/time >: 1294.68 13 | BLAS-client -o1 --transposeB 1 -m 1024 -n 1024 -k 1024 14 | BLAS kernel execution time < ns >: 723310 15 | BLAS kernel execution Gflops < 2.0*M*N*K/time >: 2968.97 16 | BLAS-client -o1 --transposeB 1 -m 2048 -n 2048 -k 2048 17 | BLAS kernel execution time < ns >: 5.6569e+06 18 | BLAS kernel execution Gflops < 2.0*M*N*K/time >: 3036.98 19 | BLAS-client -o1 --transposeB 1 -m 4096 -n 4096 -k 4096 20 | BLAS kernel execution time < ns >: 6.99098e+07 21 | BLAS kernel execution Gflops < 2.0*M*N*K/time >: 1965.95 22 | BLAS-client -o1 --transposeB 1 -m 512 -n 512 -k 512 23 | BLAS kernel execution time < ns >: 266880 24 | BLAS kernel execution Gflops < 2.0*M*N*K/time >: 1005.83 25 | BLAS-client -o1 --transposeB 1 -m 1024 -n 1024 -k 1024 26 | BLAS kernel execution time < ns >: 719337 27 | BLAS kernel execution Gflops < 2.0*M*N*K/time >: 2985.37 28 | BLAS-client -o1 --transposeB 1 -m 1536 -n 1536 -k 1536 29 | BLAS kernel execution time < ns >: 2.08752e+06 30 | BLAS kernel execution Gflops < 2.0*M*N*K/time >: 3471.94 31 | BLAS-client -o1 --transposeB 1 -m 2048 -n 2048 -k 2048 32 | BLAS kernel execution time < ns >: 5.7503e+06 33 | BLAS kernel execution Gflops < 2.0*M*N*K/time >: 2987.65 34 | BLAS-client -o1 --transposeB 1 -m 2560 -n 2560 -k 2560 35 | BLAS kernel execution time < ns >: 1.12787e+07 36 | BLAS kernel execution Gflops < 2.0*M*N*K/time >: 2975.02 37 | BLAS-client -o1 --transposeB 1 -m 3072 -n 3072 -k 3072 38 | BLAS kernel execution time < ns >: 1.72154e+07 39 | BLAS kernel execution Gflops < 2.0*M*N*K/time >: 3368.03 40 | BLAS-client -o1 --transposeB 1 -m 3584 -n 3584 -k 3584 41 | BLAS kernel execution time < ns >: 3.10087e+07 42 | BLAS kernel execution Gflops < 2.0*M*N*K/time >: 2969.28 43 | BLAS-client -o1 --transposeB 1 -m 4096 -n 4096 -k 4096 44 | BLAS kernel execution time < ns >: 6.52492e+07 45 | BLAS kernel execution Gflops < 2.0*M*N*K/time >: 2106.37 46 | BLAS-client -o1 --transposeB 1 -m 4608 -n 4608 -k 4608 47 | BLAS kernel execution time < ns >: 6.04398e+07 48 | BLAS kernel execution Gflops < 2.0*M*N*K/time >: 3237.76 49 | BLAS-client -o1 --transposeB 1 -m 5120 -n 5120 -k 5120 50 | BLAS kernel execution time < ns >: 1.10235e+08 51 | BLAS kernel execution Gflops < 2.0*M*N*K/time >: 2435.12 52 | BLAS-client -o1 --transposeB 1 -m 5632 -n 5632 -k 5632 53 | BLAS kernel execution time < ns >: 1.24523e+08 54 | BLAS kernel execution Gflops < 2.0*M*N*K/time >: 2869.24 55 | BLAS-client -o1 --transposeB 1 -m 6144 -n 6144 -k 6144 56 | BLAS kernel execution time < ns >: 2.41213e+08 57 | BLAS kernel execution Gflops < 2.0*M*N*K/time >: 1923.02 58 | -------------------------------------------------------------------------------- /benchmark/i7-4790K_OPENCL/sk_benchmark/process_forward_layers_timings.csv: -------------------------------------------------------------------------------- 1 | 0;data;0.0184396;0 2 | 1;silence;0.0036626;0 3 | 2;conv1;22.190742;699388656 4 | 3;relu1;2.6799646;0 5 | 4;pool1;14.6533894;0 6 | 5;conv2;213.1595886;14062669312 7 | 6;relu2;4.9413424;0 8 | 7;pool2;40.0590094;0 9 | 8;conv3;198.1541316;18401596416 10 | 9;relu3;5.8513244;0 11 | 10;pool3;57.0788916;0 12 | 11;ip1;3612.29921;644228317184 13 | 12;relu4;13.5888408;0 14 | 13;ip2;150.0283782;17171480576 15 | 14;relu5;5.9017318;0 16 | 15;ip3;56.2488972;33521664 17 | 16;prob;0.4782932;0 18 | -------------------------------------------------------------------------------- /benchmark/i7-4790K_OPENCL/sk_benchmark/process_memory_usage.csv: -------------------------------------------------------------------------------- 1 | Peak memory usage;2892571592 2 | data;629292 3 | label;4 4 | conv1;9547968 5 | pool1;9462528 6 | conv2;23447552 7 | pool2;23011328 8 | conv3;31961088 9 | pool3;30720000 10 | ip1;67108864 11 | ip2;33554432 12 | ip3;131072 13 | prob;131072 14 | -------------------------------------------------------------------------------- /benchmark/i7-4790K_OPENCL/sk_benchmark/process_total_timings.csv: -------------------------------------------------------------------------------- 1 | Forward;4675.582105 2 | -------------------------------------------------------------------------------- /benchmark/i7-4790K_OPENCL/sk_benchmark/train_backward_layers_timings.csv: -------------------------------------------------------------------------------- 1 | 0;label;0.0029102;0 2 | 1;data;0.003008;0 3 | 2;silence;0.0040756;0 4 | 3;conv1;29.460228;0 5 | 4;relu1;2.199109;0 6 | 5;pool1;14.1824136;0 7 | 6;conv2;334.5710306;0 8 | 7;relu2;5.581726;0 9 | 8;pool2;30.0692674;0 10 | 9;conv3;328.6651608;0 11 | 10;relu3;7.5785388;0 12 | 11;pool3;36.6120784;0 13 | 12;ip1;6340.919085;0 14 | 13;relu4;18.1010066;0 15 | 14;ip2;296.9252264;0 16 | 15;relu5;8.1503498;0 17 | 16;ip3;114.5133774;0 18 | 17;loss;0.3753408;0 19 | -------------------------------------------------------------------------------- /benchmark/i7-4790K_OPENCL/sk_benchmark/train_forward_layers_timings.csv: -------------------------------------------------------------------------------- 1 | 0;label;0.0284308;0 2 | 1;data;0.0069818;0 3 | 2;silence;0.0037074;0 4 | 3;conv1;18.3331938;699388656 5 | 4;relu1;2.5207114;0 6 | 5;pool1;14.0784632;0 7 | 6;conv2;204.5347088;14062669312 8 | 7;relu2;4.713152;0 9 | 8;pool2;39.007587;0 10 | 9;conv3;198.499838;18401596416 11 | 10;relu3;5.4154078;0 12 | 11;pool3;49.6407652;0 13 | 12;ip1;3562.445007;644228317184 14 | 13;relu4;11.6346604;0 15 | 14;ip2;150.9750644;17171480576 16 | 15;relu5;5.546087;0 17 | 16;ip3;54.8629552;33521664 18 | 17;loss;1.11714;0 19 | -------------------------------------------------------------------------------- /benchmark/i7-4790K_OPENCL/sk_benchmark/train_memory_usage.csv: -------------------------------------------------------------------------------- 1 | Peak memory usage;3204621764 2 | label;65536 3 | labeli;4 4 | data;629292 5 | datai;4 6 | conv1;9547968 7 | pool1;9462528 8 | conv2;23447552 9 | pool2;23011328 10 | conv3;31961088 11 | pool3;30720000 12 | ip1;67108864 13 | ip2;33554432 14 | ip3;131072 15 | (automatic);4 16 | -------------------------------------------------------------------------------- /benchmark/i7-4790K_OPENCL/sk_benchmark/train_total_timings.csv: -------------------------------------------------------------------------------- 1 | Forward;4483.265315 2 | Backward;7956.081515 3 | -------------------------------------------------------------------------------- /benchmark/i7-4790K_OPENCL/u_benchmark/process_forward_layers_timings.csv: -------------------------------------------------------------------------------- 1 | 0;data;0.0274634;0 2 | 1;silence;0.0033362;0 3 | 2;conv1;39.107561;1102060800 4 | 3;relu1;22.5740144;0 5 | 4;conv2;611.0325036;23765774336 6 | 5;relu2;16.6983694;0 7 | 6;conv2_relu2_0_split;0.0124272;0 8 | 7;pool1;48.1928618;0 9 | 8;conv3;119.5044252;11716111872 10 | 9;relu3;5.9813702;0 11 | 10;conv4;338.209847;23111065600 12 | 11;relu4;7.198351;0 13 | 12;conv4_relu4_0_split;0.0116692;0 14 | 13;pool2;11.8258134;0 15 | 14;conv5;93.3263156;11227732992 16 | 15;relu5;2.5349958;0 17 | 16;conv6;172.756974;21814034432 18 | 17;relu6;3.2135516;0 19 | 18;conv6_relu6_0_split;0.0128082;0 20 | 19;pool3;7.1107608;0 21 | 20;conv7;49.8762422;10274863104 22 | 21;relu7;1.3154076;0 23 | 22;conv8;88.6973652;19325255680 24 | 23;relu8;1.1421462;0 25 | 24;conv8_relu8_0_split;0.011487;0 26 | 25;pool4;3.1834572;0 27 | 26;conv9;30.6472956;8492544000 28 | 27;relu9;0.5287776;0 29 | 28;conv10;53.2272548;14796701696 30 | 29;relu10;0.4309022;0 31 | 30;upconv1;94.0610546;0 32 | 31;conv11;16.6592562;3286728704 33 | 32;mergecrop1;34.5936008;0 34 | 33;conv12;132.7075866;27517335552 35 | 34;relu11;0.893465;0 36 | 35;conv13;57.6091958;12757688320 37 | 36;relu12;0.851825;0 38 | 37;upconv2;99.516944;0 39 | 38;conv14;19.38658;2832580608 40 | 39;mergecrop2;49.8755802;0 41 | 40;conv15;210.8316242;24543452160 42 | 41;relu13;1.7811438;0 43 | 42;conv16;89.0637614;11793920000 44 | 43;relu14;1.45692;0 45 | 44;upconv3;120.6291338;0 46 | 45;conv17;22.0084344;2616320000 47 | 46;mergecrop3;74.041791;0 48 | 47;conv18;271.451327;23118441984 49 | 48;relu15;3.2541624;0 50 | 49;conv19;160.4096518;11324422144 51 | 50;relu16;3.2567222;0 52 | 51;upconv4;171.6589058;0 53 | 52;conv20;27.4826982;2507796480 54 | 53;mergecrop4;110.5378062;0 55 | 54;conv21;419.5214016;22418323200 56 | 55;relu17;7.6071614;0 57 | 56;conv22;267.7547664;11089673216 58 | 57;relu18;6.170966;0 59 | 58;ip1;21.0980888;38238176 60 | 59;prob;3.9202252;0 61 | -------------------------------------------------------------------------------- /benchmark/i7-4790K_OPENCL/u_benchmark/process_memory_usage.csv: -------------------------------------------------------------------------------- 1 | Peak memory usage;1793039512 2 | data;3926208 3 | datai;4 4 | conv1;83174400 5 | conv2;82591744 6 | conv2_relu2_0_split_0;82591744 7 | conv2_relu2_0_split_1;82591744 8 | pool1;20647936 9 | conv3;40716288 10 | conv4;40140800 11 | conv4_relu4_0_split_0;40140800 12 | conv4_relu4_0_split_1;40140800 13 | pool2;10035200 14 | conv5;19501056 15 | conv6;18939904 16 | conv6_relu6_0_split_0;18939904 17 | conv6_relu6_0_split_1;18939904 18 | pool3;4734976 19 | conv7;8921088 20 | conv8;8388608 21 | conv8_relu8_0_split_0;8388608 22 | conv8_relu8_0_split_1;8388608 23 | pool4;2097152 24 | conv9;3686400 25 | conv10;3211264 26 | upconv1;12845056 27 | conv11;6422528 28 | mergecrop1;12845056 29 | conv12;5971968 30 | conv13;5537792 31 | upconv2;22151168 32 | conv14;11075584 33 | mergecrop2;22151168 34 | conv15;10653696 35 | conv16;10240000 36 | upconv3;40960000 37 | conv17;20480000 38 | mergecrop3;40960000 39 | conv18;20072448 40 | conv19;19668992 41 | upconv4;78675968 42 | conv20;39337984 43 | mergecrop4;78675968 44 | conv21;38937600 45 | conv22;38539264 46 | ip1;1204352 47 | prob;1204352 48 | -------------------------------------------------------------------------------- /benchmark/i7-4790K_OPENCL/u_benchmark/process_total_timings.csv: -------------------------------------------------------------------------------- 1 | Forward;4604.896615 2 | -------------------------------------------------------------------------------- /benchmark/i7-4790K_OPENCL/u_benchmark/train_backward_layers_timings.csv: -------------------------------------------------------------------------------- 1 | 0;label;0.004054;0 2 | 1;data;0.0024578;0 3 | 2;silence;0.0043998;0 4 | 3;conv1;84.6204514;0 5 | 4;relu1;38.5630076;0 6 | 5;conv2;1195.344379;0 7 | 6;relu2;33.3240644;0 8 | 7;conv2_relu2_0_split;38.8607792;0 9 | 8;pool1;158.6526464;0 10 | 9;conv3;340.489479;0 11 | 10;relu3;11.0213116;0 12 | 11;conv4;624.7077012;0 13 | 12;relu4;13.5890722;0 14 | 13;conv4_relu4_0_split;19.637882;0 15 | 14;pool2;84.725778;0 16 | 15;conv5;204.9364416;0 17 | 16;relu5;7.4160284;0 18 | 17;conv6;389.9348502;0 19 | 18;relu6;7.2235968;0 20 | 19;conv6_relu6_0_split;11.132718;0 21 | 20;pool3;55.7367018;0 22 | 21;conv7;122.9356484;0 23 | 22;relu7;5.1871906;0 24 | 23;conv8;234.3254132;0 25 | 24;relu8;3.178703;0 26 | 25;conv8_relu8_0_split;6.319313;0 27 | 26;pool4;28.0772692;0 28 | 27;conv9;80.3139864;0 29 | 28;relu9;1.1405114;0 30 | 29;conv10;157.1086766;0 31 | 30;relu10;0.6114416;0 32 | 31;upconv1;78.8917476;0 33 | 32;conv11;29.9117746;0 34 | 33;mergecrop1;19.1957514;0 35 | 34;conv12;346.4667494;0 36 | 35;relu11;2.44084;0 37 | 36;conv13;157.1371;0 38 | 37;relu12;1.6447922;0 39 | 38;upconv2;58.868087;0 40 | 39;conv14;32.729889;0 41 | 40;mergecrop2;28.1414034;0 42 | 41;conv15;382.8025028;0 43 | 42;relu13;3.4846414;0 44 | 43;conv16;181.7308804;0 45 | 44;relu14;2.9267902;0 46 | 45;upconv3;50.1304372;0 47 | 46;conv17;38.1024598;0 48 | 47;mergecrop3;46.0241078;0 49 | 48;conv18;609.7237076;0 50 | 49;relu15;5.5932814;0 51 | 50;conv19;285.036058;0 52 | 51;relu16;5.949447;0 53 | 52;upconv4;84.2867804;0 54 | 53;conv20;46.5766938;0 55 | 54;mergecrop4;91.4162548;0 56 | 55;conv21;1108.95154;0 57 | 56;relu17;12.2821238;0 58 | 57;conv22;580.9277184;0 59 | 58;relu18;21.2666082;0 60 | 59;ip1;29.0947108;0 61 | 60;loss;3.8388814;0 62 | -------------------------------------------------------------------------------- /benchmark/i7-4790K_OPENCL/u_benchmark/train_forward_layers_timings.csv: -------------------------------------------------------------------------------- 1 | 0;label;0.0313716;0 2 | 1;data;0.0067282;0 3 | 2;silence;0.0042746;0 4 | 3;conv1;45.8495816;1102060800 5 | 4;relu1;32.5950046;0 6 | 5;conv2;721.7618956;23765774336 7 | 6;relu2;27.0231456;0 8 | 7;conv2_relu2_0_split;0.0120688;0 9 | 8;pool1;49.8881606;0 10 | 9;conv3;147.7298004;11716111872 11 | 10;relu3;10.561007;0 12 | 11;conv4;441.012431;23111065600 13 | 12;relu4;9.5151816;0 14 | 13;conv4_relu4_0_split;0.0112528;0 15 | 14;pool2;22.4601168;0 16 | 15;conv5;121.067904;11227732992 17 | 16;relu5;3.3821928;0 18 | 17;conv6;230.365779;21814034432 19 | 18;relu6;3.6215124;0 20 | 19;conv6_relu6_0_split;0.0163948;0 21 | 20;pool3;7.3676682;0 22 | 21;conv7;64.857858;10274863104 23 | 22;relu7;1.4240718;0 24 | 23;conv8;116.1438326;19325255680 25 | 24;relu8;1.1845462;0 26 | 25;conv8_relu8_0_split;0.0117484;0 27 | 26;pool4;3.5646118;0 28 | 27;conv9;54.2426358;8492544000 29 | 28;relu9;0.6559158;0 30 | 29;conv10;83.0037966;14796701696 31 | 30;relu10;0.4468794;0 32 | 31;upconv1;96.6003398;0 33 | 32;conv11;21.052009;3286728704 34 | 33;mergecrop1;32.4953074;0 35 | 34;conv12;167.562008;27517335552 36 | 35;relu11;2.066091;0 37 | 36;conv13;63.6825812;12757688320 38 | 37;relu12;0.7584768;0 39 | 38;upconv2;90.4547676;0 40 | 39;conv14;19.3913462;2832580608 41 | 40;mergecrop2;58.555659;0 42 | 41;conv15;254.0227314;24543452160 43 | 42;relu13;1.513095;0 44 | 43;conv16;101.9524098;11793920000 45 | 44;relu14;1.6835296;0 46 | 45;upconv3;133.1183258;0 47 | 46;conv17;24.5912018;2616320000 48 | 47;mergecrop3;78.340638;0 49 | 48;conv18;319.4565938;23118441984 50 | 49;relu15;3.7993186;0 51 | 50;conv19;182.8795482;11324422144 52 | 51;relu16;4.3304228;0 53 | 52;upconv4;213.4746864;0 54 | 53;conv20;28.6287736;2507796480 55 | 54;mergecrop4;126.2643332;0 56 | 55;conv21;557.3930256;22418323200 57 | 56;relu17;9.3335464;0 58 | 57;conv22;362.0934896;11089673216 59 | 58;relu18;9.1193492;0 60 | 59;ip1;23.1702116;38238176 61 | 60;loss;5.8316076;0 62 | -------------------------------------------------------------------------------- /benchmark/i7-4790K_OPENCL/u_benchmark/train_memory_usage.csv: -------------------------------------------------------------------------------- 1 | Peak memory usage;3098889896 2 | label;602176 3 | labeli;4 4 | data;3926208 5 | datai;4 6 | conv1;83174400 7 | conv2;82591744 8 | conv2_relu2_0_split_0;82591744 9 | conv2_relu2_0_split_1;82591744 10 | pool1;20647936 11 | conv3;40716288 12 | conv4;40140800 13 | conv4_relu4_0_split_0;40140800 14 | conv4_relu4_0_split_1;40140800 15 | pool2;10035200 16 | conv5;19501056 17 | conv6;18939904 18 | conv6_relu6_0_split_0;18939904 19 | conv6_relu6_0_split_1;18939904 20 | pool3;4734976 21 | conv7;8921088 22 | conv8;8388608 23 | conv8_relu8_0_split_0;8388608 24 | conv8_relu8_0_split_1;8388608 25 | pool4;2097152 26 | conv9;3686400 27 | conv10;3211264 28 | upconv1;12845056 29 | conv11;6422528 30 | mergecrop1;12845056 31 | conv12;5971968 32 | conv13;5537792 33 | upconv2;22151168 34 | conv14;11075584 35 | mergecrop2;22151168 36 | conv15;10653696 37 | conv16;10240000 38 | upconv3;40960000 39 | conv17;20480000 40 | mergecrop3;40960000 41 | conv18;20072448 42 | conv19;19668992 43 | upconv4;78675968 44 | conv20;39337984 45 | mergecrop4;78675968 46 | conv21;38937600 47 | conv22;38539264 48 | ip1;1204352 49 | (automatic);4 50 | -------------------------------------------------------------------------------- /benchmark/i7-4790K_OPENCL/u_benchmark/train_total_timings.csv: -------------------------------------------------------------------------------- 1 | Forward;4424.887664 2 | Backward;6702.328755 3 | -------------------------------------------------------------------------------- /benchmark/i7-4790K_OPENCL/usk_benchmark/process_forward_layers_timings.csv: -------------------------------------------------------------------------------- 1 | 0;data;0.0310962;0 2 | 1;silence;0.003935;0 3 | 2;conv1;22.9776466;638903552 4 | 3;relu1;10.691161;0 5 | 4;conv2;333.7864692;13747470336 6 | 5;relu2;9.0678912;0 7 | 6;conv2_relu2_0_split;0.0119234;0 8 | 7;pool1;21.5963518;0 9 | 8;conv3;437.042976;26253855616 10 | 9;relu3;4.081047;0 11 | 10;pool2;33.5919994;0 12 | 11;conv4;338.9476146;21813442560 13 | 12;relu4;3.3481248;0 14 | 13;pool3;29.3842186;0 15 | 14;conv5;282.9458806;18922176000 16 | 15;relu5;3.3069316;0 17 | 16;pool4;23.7714168;0 18 | 17;ip1;995.106066;141758822400 19 | 18;relu6;4.8087888;0 20 | 19;ip2;20.2963214;4425907200 21 | 20;relu7;2.430759;0 22 | 21;upconv1;155.6091418;0 23 | 22;conv6;27.4694206;4421580800 24 | 23;mergecrop1;80.9440986;0 25 | 24;conv7;333.0637032;29437263360 26 | 25;relu8;5.9509678;0 27 | 26;conv8;256.990782;9659482112 28 | 27;relu9;3.1498494;0 29 | 28;ip3;8.9535774;16646144 30 | 29;prob;1.3205736;0 31 | -------------------------------------------------------------------------------- /benchmark/i7-4790K_OPENCL/usk_benchmark/process_memory_usage.csv: -------------------------------------------------------------------------------- 1 | Peak memory usage;1145971772 2 | data;2281152 3 | datai;4 4 | conv1;48219136 5 | conv2;47775744 6 | conv2_relu2_0_split_0;47775744 7 | conv2_relu2_0_split_1;47775744 8 | pool1;11943936 9 | conv3;22794752 10 | pool2;22579200 11 | conv4;21307392 12 | pool3;20891648 13 | conv5;18483200 14 | pool4;17713152 15 | ip1;34611200 16 | ip2;17305600 17 | upconv1;69222400 18 | conv6;34611200 19 | mergecrop1;51916800 20 | conv7;34080768 21 | conv8;16777216 22 | ip3;524288 23 | prob;524288 24 | -------------------------------------------------------------------------------- /benchmark/i7-4790K_OPENCL/usk_benchmark/process_total_timings.csv: -------------------------------------------------------------------------------- 1 | Forward;3375.567904 2 | -------------------------------------------------------------------------------- /benchmark/train_process_sk_2.prototxt: -------------------------------------------------------------------------------- 1 | train { 2 | solver: "../net_sk_2out/neuraltissue_solver.prototxt" 3 | input { 4 | padding_size: 102 5 | patch_size: 128 6 | channels: 3 7 | labels: 2 8 | batch_size: 1 9 | raw_images: "train/raw" 10 | label_images: "train/labels" 11 | preprocessor { 12 | normalization: true 13 | rotation: true 14 | mirror: true 15 | clahe { 16 | clip: 4.0 17 | } 18 | crop { 19 | imagecrop: 1 20 | labelcrop: 0 21 | } 22 | blur { 23 | mean: 0.0 24 | std: 0.1 25 | ksize: 5 26 | } 27 | histeq { 28 | patch_prior: false 29 | masking: false 30 | } 31 | } 32 | } 33 | } 34 | 35 | process { 36 | process_net: "../net_sk_2out/neuraltissue_net.prototxt" 37 | input { 38 | padding_size: 102 39 | patch_size: 128 40 | channels: 3 41 | labels: 2 42 | batch_size: 1 43 | raw_images: "input" 44 | preprocessor { 45 | normalization: true 46 | clahe { 47 | clip: 4.0 48 | } 49 | crop { 50 | imagecrop: 1 51 | labelcrop: 0 52 | } 53 | } 54 | } 55 | filter_output { 56 | output_filters: false 57 | output: "sk_filters" 58 | } 59 | output { 60 | format: "tif" 61 | fp32_out: false 62 | output: "output" 63 | } 64 | } 65 | 66 | benchmark { 67 | train_index: 0 68 | process_index: 0 69 | output: "sk_benchmark" 70 | warmup_runs: 2 71 | bench_runs: 5 72 | } 73 | -------------------------------------------------------------------------------- /benchmark/train_process_u_2.prototxt: -------------------------------------------------------------------------------- 1 | train { 2 | solver: "../net_u_2out/neuraltissue_solver.prototxt" 3 | input { 4 | padding_size: 184 5 | patch_size: 388 6 | channels: 3 7 | labels: 2 8 | batch_size: 1 9 | raw_images: "train/raw" 10 | label_images: "train/labels" 11 | preprocessor { 12 | normalization: true 13 | rotation: true 14 | mirror: true 15 | clahe { 16 | clip: 4.0 17 | } 18 | crop { 19 | imagecrop: 0 20 | labelcrop: 0 21 | } 22 | blur { 23 | mean: 0.0 24 | std: 0.1 25 | ksize: 5 26 | } 27 | histeq { 28 | patch_prior: false 29 | masking: false 30 | } 31 | } 32 | } 33 | } 34 | 35 | process { 36 | process_net: "../net_u_2out/neuraltissue_net.prototxt" 37 | input { 38 | padding_size: 184 39 | patch_size: 388 40 | channels: 3 41 | labels: 2 42 | batch_size: 1 43 | raw_images: "input" 44 | preprocessor { 45 | normalization: true 46 | clahe { 47 | clip: 4.0 48 | } 49 | crop { 50 | imagecrop: 0 51 | labelcrop: 0 52 | } 53 | } 54 | } 55 | filter_output { 56 | output_filters: false 57 | output: "u_filters" 58 | } 59 | output { 60 | format: "tif" 61 | fp32_out: false 62 | output: "output" 63 | } 64 | } 65 | 66 | benchmark { 67 | train_index: 0 68 | process_index: 0 69 | output: "u_benchmark" 70 | warmup_runs: 2 71 | bench_runs: 5 72 | } 73 | -------------------------------------------------------------------------------- /benchmark/train_process_u_2small.prototxt: -------------------------------------------------------------------------------- 1 | train { 2 | solver: "../net_u_small/neuraltissue_solver.prototxt" 3 | input { 4 | padding_size: 184 5 | patch_size: 196 6 | channels: 3 7 | labels: 2 8 | batch_size: 1 9 | raw_images: "train/raw" 10 | label_images: "train/labels" 11 | preprocessor { 12 | normalization: true 13 | rotation: true 14 | mirror: true 15 | clahe { 16 | clip: 4.0 17 | } 18 | crop { 19 | imagecrop: 0 20 | labelcrop: 0 21 | } 22 | blur { 23 | mean: 0.0 24 | std: 0.1 25 | ksize: 5 26 | } 27 | histeq { 28 | patch_prior: false 29 | masking: false 30 | } 31 | } 32 | } 33 | } 34 | 35 | process { 36 | process_net: "../net_u_small/neuraltissue_net.prototxt" 37 | input { 38 | padding_size: 184 39 | patch_size: 196 40 | channels: 3 41 | labels: 2 42 | batch_size: 1 43 | raw_images: "input" 44 | preprocessor { 45 | normalization: true 46 | clahe { 47 | clip: 4.0 48 | } 49 | crop { 50 | imagecrop: 0 51 | labelcrop: 0 52 | } 53 | } 54 | } 55 | filter_output { 56 | output_filters: false 57 | output: "u_filters" 58 | } 59 | output { 60 | format: "tif" 61 | fp32_out: false 62 | output: "output" 63 | } 64 | } 65 | 66 | benchmark { 67 | train_index: 0 68 | process_index: 0 69 | output: "u_benchmark" 70 | warmup_runs: 2 71 | bench_runs: 5 72 | } 73 | -------------------------------------------------------------------------------- /benchmark/train_process_usk_2.prototxt: -------------------------------------------------------------------------------- 1 | train { 2 | solver: "../net_usk_2out/neuraltissue_solver.prototxt" 3 | input { 4 | padding_size: 180 5 | patch_size: 256 6 | channels: 3 7 | labels: 2 8 | batch_size: 1 9 | raw_images: "train/raw" 10 | label_images: "train/labels" 11 | preprocessor { 12 | normalization: true 13 | rotation: true 14 | mirror: true 15 | clahe { 16 | clip: 4.0 17 | } 18 | crop { 19 | imagecrop: 0 20 | labelcrop: 0 21 | } 22 | blur { 23 | mean: 0.0 24 | std: 0.1 25 | ksize: 5 26 | } 27 | histeq { 28 | patch_prior: false 29 | masking: false 30 | } 31 | } 32 | } 33 | } 34 | 35 | process { 36 | process_net: "../net_usk_2out/neuraltissue_net.prototxt" 37 | input { 38 | padding_size: 180 39 | patch_size: 256 40 | channels: 3 41 | labels: 2 42 | batch_size: 1 43 | raw_images: "input" 44 | preprocessor { 45 | normalization: true 46 | clahe { 47 | clip: 4.0 48 | } 49 | crop { 50 | imagecrop: 0 51 | labelcrop: 0 52 | } 53 | } 54 | } 55 | filter_output { 56 | output_filters: false 57 | output: "usk_filters" 58 | } 59 | output { 60 | format: "tif" 61 | fp32_out: false 62 | output: "output" 63 | } 64 | } 65 | 66 | benchmark { 67 | train_index: 0 68 | process_index: 0 69 | output: "usk_benchmark" 70 | warmup_runs: 2 71 | bench_runs: 5 72 | } 73 | -------------------------------------------------------------------------------- /benchmark/u_benchmark/process_forward_layers_timings.csv: -------------------------------------------------------------------------------- 1 | 0;data;0.0090552;0 2 | 1;silence;0.0038782;0 3 | 2;Convolution1;10.0284528;484662528 4 | 3;ReLU1;3.0279008;0 5 | 4;Convolution2;72.2204378;10414321664 6 | 5;ReLU2;3.0005238;0 7 | 6;Convolution2_ReLU2_0_split;0.00723;0 8 | 7;Pooling1;8.1683908;0 9 | 8;Convolution3;26.3916874;5096959488 10 | 9;ReLU3;1.4908118;0 11 | 10;Convolution4;48.441133;9980207104 12 | 11;ReLU4;1.4579302;0 13 | 12;Convolution4_ReLU4_0_split;0.0040838;0 14 | 13;Pooling2;3.9168062;0 15 | 14;Convolution5;19.0256108;4775500800 16 | 15;ReLU5;0.7177118;0 17 | 16;Convolution6;34.8015366;9133211648 18 | 17;ReLU6;0.690141;0 19 | 18;Convolution6_ReLU6_0_split;0.0032752;0 20 | 19;Pooling3;1.8104738;0 21 | 20;Convolution7;14.2014374;4160894976 22 | 21;ReLU7;0.3332222;0 23 | 22;Convolution8;24.78495;7548928000 24 | 23;ReLU8;0.307989;0 25 | 24;Convolution8_ReLU8_0_split;0.0030884;0 26 | 25;Pooling4;0.7642218;0 27 | 26;Convolution9;10.964097;3057315840 28 | 27;ReLU9;0.1422286;0 29 | 28;Convolution10;14.6336422;4831576064 30 | 29;ReLU10;0.116546;0 31 | 30;Deconvolution1;10.9683482;0 32 | 31;Convolution11;3.326726;1073217536 33 | 32;MergeCrop1;8.0441928;0 34 | 33;Convolution12;29.2545192;8493004800 35 | 34;ReLU11;0.2022996;0 36 | 35;Convolution13;12.827152;3698974720 37 | 36;ReLU12;0.1607844;0 38 | 37;Deconvolution2;13.3849872;0 39 | 38;Convolution14;2.8390096;821280768 40 | 39;MergeCrop2;12.302372;0 41 | 40;Convolution15;26.003882;6878960640 42 | 41;ReLU13;0.2852658;0 43 | 42;Convolution16;12.4286394;3189075968 44 | 43;ReLU14;0.2586616;0 45 | 44;Deconvolution3;21.1402356;0 46 | 45;Convolution17;2.8520432;707452928 47 | 46;MergeCrop3;21.144977;0 48 | 47;Convolution18;29.163433;6135197184 49 | 48;ReLU15;0.4828196;0 50 | 49;Convolution19;14.4890398;2947840000 51 | 50;ReLU16;0.452281;0 52 | 51;Deconvolution4;37.9232498;0 53 | 52;Convolution20;3.392949;652800000 54 | 53;MergeCrop4;39.0993832;0 55 | 54;Convolution21;38.8368222;5778355968 56 | 55;ReLU17;0.8638632;0 57 | 56;Convolution22;19.7834288;2829876224 58 | 57;ReLU18;0.8479158;0 59 | 58;Convolution23;0.9066796;9757664 60 | 59;prob;0.591295;0 61 | -------------------------------------------------------------------------------- /benchmark/u_benchmark/process_memory_usage.csv: -------------------------------------------------------------------------------- 1 | Peak memory usage;821678040 2 | data;1732800 3 | datai;4 4 | Convolution1;36578304 5 | Convolution2;36192256 6 | Convolution2_ReLU2_0_split_0;36192256 7 | Convolution2_ReLU2_0_split_1;36192256 8 | Pooling1;9048064 9 | Convolution3;17713152 10 | Convolution4;17334272 11 | Convolution4_ReLU4_0_split_0;17334272 12 | Convolution4_ReLU4_0_split_1;17334272 13 | Pooling2;4333568 14 | Convolution5;8294400 15 | Convolution6;7929856 16 | Convolution6_ReLU6_0_split_0;7929856 17 | Convolution6_ReLU6_0_split_1;7929856 18 | Pooling3;1982464 19 | Convolution7;3612672 20 | Convolution8;3276800 21 | Convolution8_ReLU8_0_split_0;3276800 22 | Convolution8_ReLU8_0_split_1;3276800 23 | Pooling4;819200 24 | Convolution9;1327104 25 | Convolution10;1048576 26 | Deconvolution1;4194304 27 | Convolution11;2097152 28 | MergeCrop1;4194304 29 | Convolution12;1843200 30 | Convolution13;1605632 31 | Deconvolution2;6422528 32 | Convolution14;3211264 33 | MergeCrop2;6422528 34 | Convolution15;2985984 35 | Convolution16;2768896 36 | Deconvolution3;11075584 37 | Convolution17;5537792 38 | MergeCrop3;11075584 39 | Convolution18;5326848 40 | Convolution19;5120000 41 | Deconvolution4;20480000 42 | Convolution20;10240000 43 | MergeCrop4;20480000 44 | Convolution21;10036224 45 | Convolution22;9834496 46 | Convolution23;307328 47 | prob;307328 48 | -------------------------------------------------------------------------------- /benchmark/u_benchmark/process_total_timings.csv: -------------------------------------------------------------------------------- 1 | Forward;664.611782 2 | -------------------------------------------------------------------------------- /benchmark/u_benchmark/train_backward_layers_timings.csv: -------------------------------------------------------------------------------- 1 | 0;label;0.0013398;0 2 | 1;data;0.0014006;0 3 | 2;silence;0.0110758;0 4 | 3;Convolution1;17.3956852;0 5 | 4;ReLU1;4.2096348;0 6 | 5;Convolution2;352.1860408;0 7 | 6;ReLU2;4.1510822;0 8 | 7;Convolution2_ReLU2_0_split;4.1702252;0 9 | 8;Pooling1;57.3902794;0 10 | 9;Convolution3;70.0714326;0 11 | 10;ReLU3;2.062638;0 12 | 11;Convolution4;127.9989346;0 13 | 12;ReLU4;1.999343;0 14 | 13;Convolution4_ReLU4_0_split;2.0125298;0 15 | 14;Pooling2;27.3925588;0 16 | 15;Convolution5;45.1965792;0 17 | 16;ReLU5;0.9873896;0 18 | 17;Convolution6;85.591535;0 19 | 18;ReLU6;0.9247388;0 20 | 19;Convolution6_ReLU6_0_split;0.9466976;0 21 | 20;Pooling3;12.5279514;0 22 | 21;Convolution7;33.0367912;0 23 | 22;ReLU7;0.4590788;0 24 | 23;Convolution8;58.78798;0 25 | 24;ReLU8;0.3952964;0 26 | 25;Convolution8_ReLU8_0_split;0.4054854;0 27 | 26;Pooling4;5.1896196;0 28 | 27;Convolution9;25.1609482;0 29 | 28;ReLU9;0.1943946;0 30 | 29;Convolution10;37.6213;0 31 | 30;ReLU10;0.1391888;0 32 | 31;Deconvolution1;5.899787;0 33 | 32;Convolution11;7.0467184;0 34 | 33;MergeCrop1;8.0574912;0 35 | 34;Convolution12;69.6469728;0 36 | 35;ReLU11;0.2519158;0 37 | 36;Convolution13;30.1689698;0 38 | 37;ReLU12;0.2084542;0 39 | 38;Deconvolution2;6.3535382;0 40 | 39;Convolution14;5.5931654;0 41 | 40;MergeCrop2;12.2803792;0 42 | 41;Convolution15;65.8994762;0 43 | 42;ReLU13;0.3825304;0 44 | 43;Convolution16;30.2814476;0 45 | 44;ReLU14;0.350609;0 46 | 45;Deconvolution3;8.6920224;0 47 | 46;Convolution17;5.4432254;0 48 | 47;MergeCrop3;21.1436172;0 49 | 48;Convolution18;77.936265;0 50 | 49;ReLU15;0.6520382;0 51 | 50;Convolution19;38.0886988;0 52 | 51;ReLU16;0.6248674;0 53 | 52;Deconvolution4;14.8732674;0 54 | 53;Convolution20;15.8853338;0 55 | 54;MergeCrop4;39.0823186;0 56 | 55;Convolution21;204.5975716;0 57 | 56;ReLU17;1.1903944;0 58 | 57;Convolution22;96.798616;0 59 | 58;ReLU18;1.1384794;0 60 | 59;Convolution23;3.552985;0 61 | 60;loss;0.2191624;0 62 | -------------------------------------------------------------------------------- /benchmark/u_benchmark/train_forward_layers_timings.csv: -------------------------------------------------------------------------------- 1 | 0;label;0.0095292;0 2 | 1;data;0.004297;0 3 | 2;silence;0.0035458;0 4 | 3;Convolution1;9.9730544;484662528 5 | 4;ReLU1;3.0379696;0 6 | 5;Convolution2;73.2776998;10414321664 7 | 6;ReLU2;3.019372;0 8 | 7;Convolution2_ReLU2_0_split;0.0061674;0 9 | 8;Pooling1;8.452863;0 10 | 9;Convolution3;26.7859264;5096959488 11 | 10;ReLU3;1.5012362;0 12 | 11;Convolution4;48.9759896;9980207104 13 | 12;ReLU4;1.47032;0 14 | 13;Convolution4_ReLU4_0_split;0.0064428;0 15 | 14;Pooling2;4.05094;0 16 | 15;Convolution5;19.2021876;4775500800 17 | 16;ReLU5;0.7182666;0 18 | 17;Convolution6;35.0551616;9133211648 19 | 18;ReLU6;0.69333;0 20 | 19;Convolution6_ReLU6_0_split;0.0037372;0 21 | 20;Pooling3;1.8893386;0 22 | 21;Convolution7;14.2668174;4160894976 23 | 22;ReLU7;0.3332826;0 24 | 23;Convolution8;24.9113566;7548928000 25 | 24;ReLU8;0.3109468;0 26 | 25;Convolution8_ReLU8_0_split;0.0030892;0 27 | 26;Pooling4;0.7887164;0 28 | 27;Convolution9;10.9841556;3057315840 29 | 28;ReLU9;0.1410246;0 30 | 29;Convolution10;14.6544736;4831576064 31 | 30;ReLU10;0.1219994;0 32 | 31;Deconvolution1;11.2218042;0 33 | 32;Convolution11;3.3273826;1073217536 34 | 33;MergeCrop1;8.0809738;0 35 | 34;Convolution12;29.3616996;8493004800 36 | 35;ReLU11;0.2055422;0 37 | 36;Convolution13;12.898545;3698974720 38 | 37;ReLU12;0.1617906;0 39 | 38;Deconvolution2;13.5808638;0 40 | 39;Convolution14;2.8240308;821280768 41 | 40;MergeCrop2;12.3268574;0 42 | 41;Convolution15;26.2024316;6878960640 43 | 42;ReLU13;0.2895644;0 44 | 43;Convolution16;12.523662;3189075968 45 | 44;ReLU14;0.2578004;0 46 | 45;Deconvolution3;21.4106586;0 47 | 46;Convolution17;2.8537322;707452928 48 | 47;MergeCrop3;21.231703;0 49 | 48;Convolution18;29.5882344;6135197184 50 | 49;ReLU15;0.481042;0 51 | 50;Convolution19;14.6784484;2947840000 52 | 51;ReLU16;0.4564054;0 53 | 52;Deconvolution4;38.3680954;0 54 | 53;Convolution20;3.3971234;652800000 55 | 54;MergeCrop4;39.194834;0 56 | 55;Convolution21;39.5433814;5778355968 57 | 56;ReLU17;0.8682294;0 58 | 57;Convolution22;20.1121936;2829876224 59 | 58;ReLU18;0.862644;0 60 | 59;Convolution23;0.9148318;9757664 61 | 60;loss;0.8910252;0 62 | -------------------------------------------------------------------------------- /benchmark/u_benchmark/train_memory_usage.csv: -------------------------------------------------------------------------------- 1 | Peak memory usage;1412440552 2 | label;153664 3 | labeli;4 4 | data;1732800 5 | datai;4 6 | Convolution1;36578304 7 | Convolution2;36192256 8 | Convolution2_ReLU2_0_split_0;36192256 9 | Convolution2_ReLU2_0_split_1;36192256 10 | Pooling1;9048064 11 | Convolution3;17713152 12 | Convolution4;17334272 13 | Convolution4_ReLU4_0_split_0;17334272 14 | Convolution4_ReLU4_0_split_1;17334272 15 | Pooling2;4333568 16 | Convolution5;8294400 17 | Convolution6;7929856 18 | Convolution6_ReLU6_0_split_0;7929856 19 | Convolution6_ReLU6_0_split_1;7929856 20 | Pooling3;1982464 21 | Convolution7;3612672 22 | Convolution8;3276800 23 | Convolution8_ReLU8_0_split_0;3276800 24 | Convolution8_ReLU8_0_split_1;3276800 25 | Pooling4;819200 26 | Convolution9;1327104 27 | Convolution10;1048576 28 | Deconvolution1;4194304 29 | Convolution11;2097152 30 | MergeCrop1;4194304 31 | Convolution12;1843200 32 | Convolution13;1605632 33 | Deconvolution2;6422528 34 | Convolution14;3211264 35 | MergeCrop2;6422528 36 | Convolution15;2985984 37 | Convolution16;2768896 38 | Deconvolution3;11075584 39 | Convolution17;5537792 40 | MergeCrop3;11075584 41 | Convolution18;5326848 42 | Convolution19;5120000 43 | Deconvolution4;20480000 44 | Convolution20;10240000 45 | MergeCrop4;20480000 46 | Convolution21;10036224 47 | Convolution22;9834496 48 | Convolution23;307328 49 | (automatic);4 50 | -------------------------------------------------------------------------------- /benchmark/u_benchmark/train_total_timings.csv: -------------------------------------------------------------------------------- 1 | Forward;677.8058274 2 | Backward;1775.792586 3 | -------------------------------------------------------------------------------- /dataset_01/.gitignore: -------------------------------------------------------------------------------- 1 | input 2 | output 3 | golden 4 | sk_filters 5 | usk_filters 6 | u_filters 7 | validate_output 8 | results 9 | -------------------------------------------------------------------------------- /dataset_01/process.sh: -------------------------------------------------------------------------------- 1 | ./../../caffe_neural_tool/build/caffe_neural_tool --gpu 0 --graphic --process 0 --proto 'train_process_usk_2.prototxt' 2 | -------------------------------------------------------------------------------- /dataset_01/train.sh: -------------------------------------------------------------------------------- 1 | ./../../caffe_neural_tool/build/caffe_neural_tool --gpu 1 --train 0 --proto 'train_process_sk_2.prototxt' 2 | -------------------------------------------------------------------------------- /dataset_01/train/labels/labels00000000.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/naibaf7/caffe_neural_models/9d372c4bc599029902185e19f89e5c39f842fff7/dataset_01/train/labels/labels00000000.png -------------------------------------------------------------------------------- /dataset_01/train/labels/labels00000001.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/naibaf7/caffe_neural_models/9d372c4bc599029902185e19f89e5c39f842fff7/dataset_01/train/labels/labels00000001.png -------------------------------------------------------------------------------- /dataset_01/train/labels/labels00000003.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/naibaf7/caffe_neural_models/9d372c4bc599029902185e19f89e5c39f842fff7/dataset_01/train/labels/labels00000003.png -------------------------------------------------------------------------------- /dataset_01/train/labels/labels00000004.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/naibaf7/caffe_neural_models/9d372c4bc599029902185e19f89e5c39f842fff7/dataset_01/train/labels/labels00000004.png -------------------------------------------------------------------------------- /dataset_01/train/labels/labels00000005.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/naibaf7/caffe_neural_models/9d372c4bc599029902185e19f89e5c39f842fff7/dataset_01/train/labels/labels00000005.png -------------------------------------------------------------------------------- /dataset_01/train/labels/labels00000006.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/naibaf7/caffe_neural_models/9d372c4bc599029902185e19f89e5c39f842fff7/dataset_01/train/labels/labels00000006.png -------------------------------------------------------------------------------- /dataset_01/train/labels/labels00000008.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/naibaf7/caffe_neural_models/9d372c4bc599029902185e19f89e5c39f842fff7/dataset_01/train/labels/labels00000008.png -------------------------------------------------------------------------------- /dataset_01/train/labels/labels00000009.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/naibaf7/caffe_neural_models/9d372c4bc599029902185e19f89e5c39f842fff7/dataset_01/train/labels/labels00000009.png -------------------------------------------------------------------------------- /dataset_01/train/labels/labels00000010.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/naibaf7/caffe_neural_models/9d372c4bc599029902185e19f89e5c39f842fff7/dataset_01/train/labels/labels00000010.png -------------------------------------------------------------------------------- /dataset_01/train/labels/labels00000011.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/naibaf7/caffe_neural_models/9d372c4bc599029902185e19f89e5c39f842fff7/dataset_01/train/labels/labels00000011.png -------------------------------------------------------------------------------- /dataset_01/train/labels/labels00000013.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/naibaf7/caffe_neural_models/9d372c4bc599029902185e19f89e5c39f842fff7/dataset_01/train/labels/labels00000013.png -------------------------------------------------------------------------------- /dataset_01/train/labels/labels00000014.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/naibaf7/caffe_neural_models/9d372c4bc599029902185e19f89e5c39f842fff7/dataset_01/train/labels/labels00000014.png -------------------------------------------------------------------------------- /dataset_01/train/labels/labels00000015.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/naibaf7/caffe_neural_models/9d372c4bc599029902185e19f89e5c39f842fff7/dataset_01/train/labels/labels00000015.png -------------------------------------------------------------------------------- /dataset_01/train/labels/labels00000016.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/naibaf7/caffe_neural_models/9d372c4bc599029902185e19f89e5c39f842fff7/dataset_01/train/labels/labels00000016.png -------------------------------------------------------------------------------- /dataset_01/train/labels/labels00000018.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/naibaf7/caffe_neural_models/9d372c4bc599029902185e19f89e5c39f842fff7/dataset_01/train/labels/labels00000018.png -------------------------------------------------------------------------------- /dataset_01/train/labels/labels00000019.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/naibaf7/caffe_neural_models/9d372c4bc599029902185e19f89e5c39f842fff7/dataset_01/train/labels/labels00000019.png -------------------------------------------------------------------------------- /dataset_01/train/raw/00.tif: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/naibaf7/caffe_neural_models/9d372c4bc599029902185e19f89e5c39f842fff7/dataset_01/train/raw/00.tif -------------------------------------------------------------------------------- /dataset_01/train/raw/01.tif: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/naibaf7/caffe_neural_models/9d372c4bc599029902185e19f89e5c39f842fff7/dataset_01/train/raw/01.tif -------------------------------------------------------------------------------- /dataset_01/train/raw/03.tif: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/naibaf7/caffe_neural_models/9d372c4bc599029902185e19f89e5c39f842fff7/dataset_01/train/raw/03.tif -------------------------------------------------------------------------------- /dataset_01/train/raw/04.tif: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/naibaf7/caffe_neural_models/9d372c4bc599029902185e19f89e5c39f842fff7/dataset_01/train/raw/04.tif -------------------------------------------------------------------------------- /dataset_01/train/raw/05.tif: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/naibaf7/caffe_neural_models/9d372c4bc599029902185e19f89e5c39f842fff7/dataset_01/train/raw/05.tif -------------------------------------------------------------------------------- /dataset_01/train/raw/06.tif: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/naibaf7/caffe_neural_models/9d372c4bc599029902185e19f89e5c39f842fff7/dataset_01/train/raw/06.tif -------------------------------------------------------------------------------- /dataset_01/train/raw/08.tif: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/naibaf7/caffe_neural_models/9d372c4bc599029902185e19f89e5c39f842fff7/dataset_01/train/raw/08.tif -------------------------------------------------------------------------------- /dataset_01/train/raw/09.tif: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/naibaf7/caffe_neural_models/9d372c4bc599029902185e19f89e5c39f842fff7/dataset_01/train/raw/09.tif -------------------------------------------------------------------------------- /dataset_01/train/raw/10.tif: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/naibaf7/caffe_neural_models/9d372c4bc599029902185e19f89e5c39f842fff7/dataset_01/train/raw/10.tif -------------------------------------------------------------------------------- /dataset_01/train/raw/11.tif: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/naibaf7/caffe_neural_models/9d372c4bc599029902185e19f89e5c39f842fff7/dataset_01/train/raw/11.tif -------------------------------------------------------------------------------- /dataset_01/train/raw/13.tif: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/naibaf7/caffe_neural_models/9d372c4bc599029902185e19f89e5c39f842fff7/dataset_01/train/raw/13.tif -------------------------------------------------------------------------------- /dataset_01/train/raw/14.tif: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/naibaf7/caffe_neural_models/9d372c4bc599029902185e19f89e5c39f842fff7/dataset_01/train/raw/14.tif -------------------------------------------------------------------------------- /dataset_01/train/raw/15.tif: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/naibaf7/caffe_neural_models/9d372c4bc599029902185e19f89e5c39f842fff7/dataset_01/train/raw/15.tif -------------------------------------------------------------------------------- /dataset_01/train/raw/16.tif: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/naibaf7/caffe_neural_models/9d372c4bc599029902185e19f89e5c39f842fff7/dataset_01/train/raw/16.tif -------------------------------------------------------------------------------- /dataset_01/train/raw/18.tif: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/naibaf7/caffe_neural_models/9d372c4bc599029902185e19f89e5c39f842fff7/dataset_01/train/raw/18.tif -------------------------------------------------------------------------------- /dataset_01/train/raw/19.tif: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/naibaf7/caffe_neural_models/9d372c4bc599029902185e19f89e5c39f842fff7/dataset_01/train/raw/19.tif -------------------------------------------------------------------------------- /dataset_01/train2.sh: -------------------------------------------------------------------------------- 1 | ./../../caffe_neural_tool/build/caffe_neural_tool --gpu 3 --train 0 --graphic --proto 'train_process_usk_9.prototxt' 2 | -------------------------------------------------------------------------------- /dataset_01/train_debug.sh: -------------------------------------------------------------------------------- 1 | gdb --args ../../caffe_neural_tool/build/caffe_neural_tool_dbg --gpu 0 --train 0 --graphic --proto 'train_process_sk_9.prototxt' 2 | -------------------------------------------------------------------------------- /dataset_01/train_process_sk_2.prototxt: -------------------------------------------------------------------------------- 1 | train { 2 | #solverstate: "neuraltissue_sk_2out_iter_4000.solverstate" 3 | solver: "../net_sk_2out/neuraltissue_solver.prototxt" 4 | input { 5 | padding_size: 102 6 | patch_size: 128 7 | channels: 3 8 | labels: 2 9 | batch_size: 1 10 | raw_images: "train/raw" 11 | label_images: "train/labels" 12 | preprocessor { 13 | label_consolidate { 14 | # 0 - 15 | label: 0 16 | # 1 / 17 | label: 0 18 | # 2 | 19 | label: 0 20 | # 3 \ 21 | label: 0 22 | # 4 + 23 | label: 0 24 | # 5 glia 25 | label: 0 26 | # 6 mito 27 | label: 1 28 | # 7 synapse 29 | label: 0 30 | # 8 interior 31 | label: 1 32 | } 33 | normalization: true 34 | rotation: true 35 | mirror: true 36 | clahe { 37 | clip: 4.0 38 | } 39 | crop { 40 | imagecrop: 1 41 | labelcrop: 0 42 | } 43 | blur { 44 | mean: 0.0 45 | std: 0.1 46 | ksize: 5 47 | } 48 | histeq { 49 | patch_prior: false 50 | masking: false 51 | } 52 | } 53 | } 54 | } 55 | 56 | process { 57 | process_net: "../net_sk_2out/neuraltissue_net.prototxt" 58 | caffemodel: "neuraltissue_sk_2out_iter_20000.caffemodel" 59 | input { 60 | padding_size: 102 61 | patch_size: 128 62 | channels: 3 63 | labels: 2 64 | batch_size: 1 65 | raw_images: "validate_raw" 66 | preprocessor { 67 | normalization: true 68 | clahe { 69 | clip: 4.0 70 | } 71 | crop { 72 | imagecrop: 1 73 | labelcrop: 0 74 | } 75 | } 76 | } 77 | filter_output { 78 | output_filters: false 79 | output: "sk_filters" 80 | } 81 | output { 82 | format: "tif" 83 | fp32_out: true 84 | output: "validate_output" 85 | } 86 | } 87 | -------------------------------------------------------------------------------- /dataset_01/train_process_sk_9.prototxt: -------------------------------------------------------------------------------- 1 | train { 2 | #solverstate: "neuraltissue_sk_9out_iter_16000.solverstate" 3 | solver: "../net_sk_9out/neuraltissue_solver.prototxt" 4 | input { 5 | padding_size: 102 6 | patch_size: 64 7 | channels: 3 8 | labels: 9 9 | batch_size: 1 10 | raw_images: "train/raw" 11 | label_images: "train/labels" 12 | preprocessor { 13 | normalization: true 14 | rotation: false 15 | mirror: false 16 | clahe { 17 | clip: 4.0 18 | } 19 | crop { 20 | imagecrop: 1 21 | labelcrop: 0 22 | } 23 | blur { 24 | mean: 0.0 25 | std: 0.1 26 | ksize: 5 27 | } 28 | histeq { 29 | patch_prior: true 30 | masking: true 31 | } 32 | } 33 | } 34 | } 35 | 36 | process { 37 | process_net: "../net_sk_9out/neuraltissue_net.prototxt" 38 | caffemodel: "neuraltissue_sk_9out_iter_16000.caffemodel" 39 | input { 40 | padding_size: 102 41 | patch_size: 128 42 | channels: 3 43 | labels: 9 44 | batch_size: 1 45 | raw_images: "input" 46 | preprocessor { 47 | normalization: true 48 | clahe { 49 | clip: 4.0 50 | } 51 | crop { 52 | imagecrop: 1 53 | labelcrop: 0 54 | } 55 | } 56 | } 57 | filter_output { 58 | output_filters: false 59 | output: "sk_filters" 60 | } 61 | output { 62 | format: "tif" 63 | fp32_out: false 64 | output: "output" 65 | } 66 | } 67 | -------------------------------------------------------------------------------- /dataset_01/train_process_u_2.prototxt: -------------------------------------------------------------------------------- 1 | train { 2 | # solverstate: "neuraltissue_u_2out_iter_10000.solverstate" 3 | solver: "../net_u_2out/neuraltissue_solver.prototxt" 4 | input { 5 | padding_size: 184 6 | patch_size: 388 7 | channels: 3 8 | labels: 2 9 | batch_size: 1 10 | raw_images: "train/raw" 11 | label_images: "train/labels" 12 | preprocessor { 13 | label_consolidate { 14 | # 0 - 15 | label: 0 16 | # 1 / 17 | label: 0 18 | # 2 | 19 | label: 0 20 | # 3 \ 21 | label: 0 22 | # 4 + 23 | label: 0 24 | # 5 glia 25 | label: 0 26 | # 6 mito 27 | label: 1 28 | # 7 synapse 29 | label: 0 30 | # 8 interior 31 | label: 1 32 | } 33 | normalization: true 34 | rotation: true 35 | mirror: true 36 | clahe { 37 | clip: 4.0 38 | } 39 | crop { 40 | imagecrop: 0 41 | labelcrop: 0 42 | } 43 | blur { 44 | mean: 0.0 45 | std: 0.1 46 | ksize: 5 47 | } 48 | histeq { 49 | patch_prior: false 50 | masking: false 51 | } 52 | } 53 | } 54 | } 55 | 56 | process { 57 | process_net: "../net_u_2out/neuraltissue_net.prototxt" 58 | caffemodel: "neuraltissue_u_2out_iter_20000.caffemodel" 59 | input { 60 | padding_size: 184 61 | patch_size: 388 62 | channels: 3 63 | labels: 2 64 | batch_size: 1 65 | raw_images: "validate_raw" 66 | preprocessor { 67 | normalization: true 68 | clahe { 69 | clip: 4.0 70 | } 71 | crop { 72 | imagecrop: 0 73 | labelcrop: 0 74 | } 75 | } 76 | } 77 | filter_output { 78 | output_filters: false 79 | output: "u_filters" 80 | } 81 | output { 82 | format: "tif" 83 | fp32_out: true 84 | output: "validate_output" 85 | } 86 | } 87 | -------------------------------------------------------------------------------- /dataset_01/train_process_u_9.prototxt: -------------------------------------------------------------------------------- 1 | train { 2 | solverstate: "neuraltissue_u_9out_iter_26000.solverstate" 3 | solver: "../net_u_9out/neuraltissue_solver.prototxt" 4 | input { 5 | padding_size: 184 6 | patch_size: 388 7 | channels: 3 8 | labels: 9 9 | batch_size: 1 10 | raw_images: "train/raw" 11 | label_images: "train/labels" 12 | preprocessor { 13 | normalization: true 14 | rotation: false 15 | mirror: false 16 | clahe { 17 | clip: 4.0 18 | } 19 | crop { 20 | imagecrop: 0 21 | labelcrop: 0 22 | } 23 | blur { 24 | mean: 0.0 25 | std: 0.1 26 | ksize: 5 27 | } 28 | histeq { 29 | patch_prior: true 30 | masking: true 31 | } 32 | } 33 | } 34 | } 35 | 36 | process { 37 | process_net: "../net_u_9out/neuraltissue_net.prototxt" 38 | caffemodel: "neuraltissue_u_9out_iter_26000.caffemodel" 39 | input { 40 | padding_size: 184 41 | patch_size: 388 42 | channels: 3 43 | labels: 9 44 | batch_size: 1 45 | raw_images: "input" 46 | preprocessor { 47 | normalization: true 48 | clahe { 49 | clip: 4.0 50 | } 51 | crop { 52 | imagecrop: 0 53 | labelcrop: 0 54 | } 55 | } 56 | } 57 | filter_output { 58 | output_filters: false 59 | output: "u_filters" 60 | } 61 | output { 62 | format: "tif" 63 | fp32_out: false 64 | output: "output" 65 | } 66 | } 67 | -------------------------------------------------------------------------------- /dataset_01/train_process_upsample.prototxt: -------------------------------------------------------------------------------- 1 | train { 2 | # solverstate: "neuraltissue_upsample_iter_14000.solverstate" 3 | solver: "../net_upsample/neuraltissue_solver.prototxt" 4 | input { 5 | padding_size: 0 6 | patch_size: 256 7 | channels: 3 8 | labels: 3 9 | batch_size: 1 10 | raw_images: "train/raw" 11 | label_images: "train/labels" 12 | preprocessor { 13 | normalization: true 14 | rotation: false 15 | mirror: false 16 | clahe { 17 | clip: 4.0 18 | } 19 | crop { 20 | imagecrop: 0 21 | labelcrop: 0 22 | } 23 | blur { 24 | mean: 0.0 25 | std: 0.1 26 | ksize: 5 27 | } 28 | histeq { 29 | patch_prior: false 30 | masking: false 31 | } 32 | } 33 | } 34 | } 35 | 36 | process { 37 | process_net: "../net_upsample/neuraltissue_process.prototxt" 38 | # caffemodel: "neuraltissue_usk_9out_iter_2000.caffemodel" 39 | input { 40 | padding_size: 0 41 | patch_size: 256 42 | channels: 3 43 | labels: 3 44 | batch_size: 1 45 | raw_images: "input" 46 | } 47 | filter_output { 48 | output_filters: true 49 | output: "upsample_filters" 50 | } 51 | output { 52 | format: "tif" 53 | fp32_out: false 54 | output: "output" 55 | } 56 | } 57 | -------------------------------------------------------------------------------- /dataset_01/train_process_usk_2.prototxt: -------------------------------------------------------------------------------- 1 | train { 2 | solverstate: "neuraltissue_usk_2out_iter_20000.solverstate" 3 | solver: "../net_usk_2out/neuraltissue_solver.prototxt" 4 | input { 5 | padding_size: 180 6 | patch_size: 512 7 | channels: 3 8 | labels: 2 9 | batch_size: 1 10 | raw_images: "train/raw" 11 | label_images: "train/labels" 12 | preprocessor { 13 | label_consolidate { 14 | # 0 - 15 | label: 0 16 | # 1 / 17 | label: 0 18 | # 2 | 19 | label: 0 20 | # 3 \ 21 | label: 0 22 | # 4 + 23 | label: 0 24 | # 5 glia 25 | label: 0 26 | # 6 mito 27 | label: 1 28 | # 7 synapse 29 | label: 0 30 | # 8 interior 31 | label: 1 32 | } 33 | normalization: true 34 | rotation: true 35 | mirror: true 36 | clahe { 37 | clip: 4.0 38 | } 39 | crop { 40 | imagecrop: 0 41 | labelcrop: 0 42 | } 43 | blur { 44 | mean: 0.0 45 | std: 0.1 46 | ksize: 5 47 | } 48 | histeq { 49 | patch_prior: false 50 | masking: false 51 | } 52 | } 53 | } 54 | } 55 | 56 | process { 57 | process_net: "../net_usk_2out/neuraltissue_net.prototxt" 58 | caffemodel: "neuraltissue_usk_2out_iter_20000.caffemodel" 59 | input { 60 | padding_size: 180 61 | patch_size: 512 62 | channels: 3 63 | labels: 2 64 | batch_size: 1 65 | raw_images: "validate_raw" 66 | preprocessor { 67 | normalization: true 68 | clahe { 69 | clip: 4.0 70 | } 71 | crop { 72 | imagecrop: 0 73 | labelcrop: 0 74 | } 75 | } 76 | } 77 | filter_output { 78 | output_filters: false 79 | output: "usk_filters" 80 | } 81 | output { 82 | format: "tif" 83 | fp32_out: true 84 | output: "validate_output" 85 | } 86 | } 87 | -------------------------------------------------------------------------------- /dataset_01/train_process_usk_9.prototxt: -------------------------------------------------------------------------------- 1 | train { 2 | solverstate: "neuraltissue_usk_9out_iter_10000.solverstate" 3 | solver: "../net_usk_9out/neuraltissue_solver.prototxt" 4 | input { 5 | padding_size: 180 6 | patch_size: 512 7 | channels: 3 8 | labels: 9 9 | batch_size: 1 10 | raw_images: "train/raw" 11 | label_images: "train/labels" 12 | preprocessor { 13 | normalization: true 14 | rotation: false 15 | mirror: false 16 | clahe { 17 | clip: 4.0 18 | } 19 | crop { 20 | imagecrop: 0 21 | labelcrop: 0 22 | } 23 | blur { 24 | mean: 0.0 25 | std: 0.1 26 | ksize: 5 27 | } 28 | histeq { 29 | patch_prior: true 30 | masking: true 31 | } 32 | } 33 | } 34 | } 35 | 36 | process { 37 | process_net: "../net_usk_9out/neuraltissue_net.prototxt" 38 | caffemodel: "neuraltissue_usk_9out_iter_10000.caffemodel" 39 | input { 40 | padding_size: 180 41 | patch_size: 512 42 | channels: 3 43 | labels: 9 44 | batch_size: 1 45 | raw_images: "input" 46 | preprocessor { 47 | normalization: true 48 | clahe { 49 | clip: 4.0 50 | } 51 | crop { 52 | imagecrop: 0 53 | labelcrop: 0 54 | } 55 | } 56 | } 57 | filter_output { 58 | output_filters: false 59 | output: "usk_filters" 60 | } 61 | output { 62 | format: "tif" 63 | fp32_out: false 64 | output: "output" 65 | } 66 | } 67 | -------------------------------------------------------------------------------- /dataset_01/validate_raw/validate_raw.tif: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/naibaf7/caffe_neural_models/9d372c4bc599029902185e19f89e5c39f842fff7/dataset_01/validate_raw/validate_raw.tif -------------------------------------------------------------------------------- /dataset_01/validate_target/labels00000002.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/naibaf7/caffe_neural_models/9d372c4bc599029902185e19f89e5c39f842fff7/dataset_01/validate_target/labels00000002.png -------------------------------------------------------------------------------- /dataset_01/validate_target/labels00000002_cons.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/naibaf7/caffe_neural_models/9d372c4bc599029902185e19f89e5c39f842fff7/dataset_01/validate_target/labels00000002_cons.png -------------------------------------------------------------------------------- /dataset_01/validate_target/labels00000007.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/naibaf7/caffe_neural_models/9d372c4bc599029902185e19f89e5c39f842fff7/dataset_01/validate_target/labels00000007.png -------------------------------------------------------------------------------- /dataset_01/validate_target/labels00000007_cons.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/naibaf7/caffe_neural_models/9d372c4bc599029902185e19f89e5c39f842fff7/dataset_01/validate_target/labels00000007_cons.png -------------------------------------------------------------------------------- /dataset_01/validate_target/labels00000012.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/naibaf7/caffe_neural_models/9d372c4bc599029902185e19f89e5c39f842fff7/dataset_01/validate_target/labels00000012.png -------------------------------------------------------------------------------- /dataset_01/validate_target/labels00000012_cons.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/naibaf7/caffe_neural_models/9d372c4bc599029902185e19f89e5c39f842fff7/dataset_01/validate_target/labels00000012_cons.png -------------------------------------------------------------------------------- /dataset_01/validate_target/labels00000017.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/naibaf7/caffe_neural_models/9d372c4bc599029902185e19f89e5c39f842fff7/dataset_01/validate_target/labels00000017.png -------------------------------------------------------------------------------- /dataset_01/validate_target/labels00000017_cons.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/naibaf7/caffe_neural_models/9d372c4bc599029902185e19f89e5c39f842fff7/dataset_01/validate_target/labels00000017_cons.png -------------------------------------------------------------------------------- /dataset_01/validate_target/validate_target.tif: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/naibaf7/caffe_neural_models/9d372c4bc599029902185e19f89e5c39f842fff7/dataset_01/validate_target/validate_target.tif -------------------------------------------------------------------------------- /dataset_02/.gitignore: -------------------------------------------------------------------------------- 1 | input 2 | output 3 | golden 4 | sk_filters 5 | usk_filters 6 | u_filters 7 | results 8 | -------------------------------------------------------------------------------- /dataset_02/process.sh: -------------------------------------------------------------------------------- 1 | ./../../caffe_neural_tool/build/caffe_neural_tool --gpu 1 --process 0 --proto 'train_process_usk_2.prototxt' 2 | -------------------------------------------------------------------------------- /dataset_02/train.sh: -------------------------------------------------------------------------------- 1 | ./../../caffe_neural_tool/build/caffe_neural_tool --gpu 1 --train 0 --proto 'train_process_usk_2.prototxt' 2 | -------------------------------------------------------------------------------- /dataset_02/train/labels/train-labels00.tif: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/naibaf7/caffe_neural_models/9d372c4bc599029902185e19f89e5c39f842fff7/dataset_02/train/labels/train-labels00.tif -------------------------------------------------------------------------------- /dataset_02/train/raw/train-volume00.tif: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/naibaf7/caffe_neural_models/9d372c4bc599029902185e19f89e5c39f842fff7/dataset_02/train/raw/train-volume00.tif -------------------------------------------------------------------------------- /dataset_02/train2.sh: -------------------------------------------------------------------------------- 1 | ./../../caffe_neural_tool/build/caffe_neural_tool --gpu 2 --train 0 --proto 'train_process_sk_2.prototxt' 2 | -------------------------------------------------------------------------------- /dataset_02/train_process_sk_2.prototxt: -------------------------------------------------------------------------------- 1 | train { 2 | solverstate: "neuraltissue_sk_2out_iter_10000.solverstate" 3 | solver: "../net_sk_2out/neuraltissue_solver.prototxt" 4 | input { 5 | padding_size: 102 6 | patch_size: 128 7 | channels: 3 8 | labels: 2 9 | batch_size: 1 10 | raw_images: "train/raw" 11 | label_images: "train/labels" 12 | preprocessor { 13 | normalization: true 14 | rotation: true 15 | mirror: true 16 | clahe { 17 | clip: 4.0 18 | } 19 | crop { 20 | imagecrop: 1 21 | labelcrop: 0 22 | } 23 | blur { 24 | mean: 0.0 25 | std: 0.1 26 | ksize: 5 27 | } 28 | histeq { 29 | patch_prior: false 30 | masking: false 31 | } 32 | } 33 | } 34 | } 35 | 36 | process { 37 | process_net: "../net_sk_2out/neuraltissue_net.prototxt" 38 | caffemodel: "neuraltissue_sk_2out_iter_10000.caffemodel" 39 | input { 40 | padding_size: 102 41 | patch_size: 128 42 | channels: 3 43 | labels: 2 44 | batch_size: 1 45 | raw_images: "input" 46 | preprocessor { 47 | normalization: true 48 | clahe { 49 | clip: 4.0 50 | } 51 | crop { 52 | imagecrop: 1 53 | labelcrop: 0 54 | } 55 | } 56 | } 57 | filter_output { 58 | output_filters: false 59 | output: "sk_filters" 60 | } 61 | output { 62 | format: "tif" 63 | fp32_out: true 64 | output: "output" 65 | } 66 | } 67 | -------------------------------------------------------------------------------- /dataset_02/train_process_u_2.prototxt: -------------------------------------------------------------------------------- 1 | train { 2 | # solverstate: "neuraltissue_u_2out_iter_6000.solverstate" 3 | solver: "../net_u_2out/neuraltissue_solver.prototxt" 4 | input { 5 | padding_size: 184 6 | patch_size: 388 7 | channels: 3 8 | labels: 2 9 | batch_size: 1 10 | raw_images: "train/raw" 11 | label_images: "train/labels" 12 | preprocessor { 13 | normalization: true 14 | rotation: false 15 | mirror: false 16 | clahe { 17 | clip: 4.0 18 | } 19 | crop { 20 | imagecrop: 0 21 | labelcrop: 0 22 | } 23 | blur { 24 | mean: 0.0 25 | std: 0.1 26 | ksize: 5 27 | } 28 | histeq { 29 | patch_prior: false 30 | masking: false 31 | } 32 | } 33 | } 34 | } 35 | 36 | process { 37 | process_net: "../net_u_2out/neuraltissue_net.prototxt" 38 | caffemodel: "neuraltissue_u_2out_iter_20000.caffemodel" 39 | input { 40 | padding_size: 184 41 | patch_size: 388 42 | channels: 3 43 | labels: 2 44 | batch_size: 1 45 | raw_images: "input" 46 | preprocessor { 47 | normalization: true 48 | clahe { 49 | clip: 4.0 50 | } 51 | crop { 52 | imagecrop: 0 53 | labelcrop: 0 54 | } 55 | } 56 | } 57 | filter_output { 58 | output_filters: true 59 | output: "u_filters" 60 | } 61 | output { 62 | format: "tif" 63 | fp32_out: true 64 | output: "output" 65 | } 66 | } 67 | -------------------------------------------------------------------------------- /dataset_02/train_process_usk_2.prototxt: -------------------------------------------------------------------------------- 1 | train { 2 | # solverstate: "neuraltissue_usk_2out_iter_16000.solverstate" 3 | solver: "../net_usk_2out/neuraltissue_solver.prototxt" 4 | input { 5 | padding_size: 180 6 | patch_size: 512 7 | channels: 3 8 | labels: 2 9 | batch_size: 1 10 | raw_images: "train/raw" 11 | label_images: "train/labels" 12 | preprocessor { 13 | normalization: true 14 | rotation: true 15 | mirror: true 16 | clahe { 17 | clip: 4.0 18 | } 19 | crop { 20 | imagecrop: 0 21 | labelcrop: 0 22 | } 23 | blur { 24 | mean: 0.0 25 | std: 0.1 26 | ksize: 5 27 | } 28 | histeq { 29 | patch_prior: false 30 | masking: false 31 | } 32 | } 33 | } 34 | } 35 | 36 | process { 37 | process_net: "../net_usk_2out/neuraltissue_net.prototxt" 38 | caffemodel: "neuraltissue_usk_2out_iter_10000.caffemodel" 39 | input { 40 | padding_size: 180 41 | patch_size: 512 42 | channels: 3 43 | labels: 2 44 | batch_size: 1 45 | raw_images: "input" 46 | preprocessor { 47 | normalization: true 48 | clahe { 49 | clip: 4.0 50 | } 51 | crop { 52 | imagecrop: 0 53 | labelcrop: 0 54 | } 55 | } 56 | } 57 | filter_output { 58 | output_filters: false 59 | output: "usk_filters" 60 | } 61 | output { 62 | format: "tif" 63 | fp32_out: true 64 | output: "output" 65 | } 66 | } 67 | -------------------------------------------------------------------------------- /dataset_06/.gitignore: -------------------------------------------------------------------------------- 1 | *.h5 2 | *.hdf5 3 | -------------------------------------------------------------------------------- /dataset_06/config.py: -------------------------------------------------------------------------------- 1 | # Specify the path caffe here 2 | caffe_path = "../../caffe_gt" 3 | # Specify wether or not to compile caffe 4 | library_compile = True 5 | 6 | # Specify the device to use 7 | device_id = 2 8 | 9 | # Specify the solver file 10 | solver_proto = "net/solver.prototxt" 11 | 12 | # Specify values for testing 13 | test_net = "net/net_test.prototxt" 14 | trained_model = "net__iter_12000.caffemodel" 15 | 16 | output_folder = "processed" 17 | 18 | output_dims = [44, 44, 44] 19 | input_padding = [88, 88, 88] 20 | 21 | border_reflect = False 22 | 23 | # Select "train" or "process" 24 | mode = "train" 25 | -------------------------------------------------------------------------------- /dataset_06/config.pyc: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/naibaf7/caffe_neural_models/9d372c4bc599029902185e19f89e5c39f842fff7/dataset_06/config.pyc -------------------------------------------------------------------------------- /dataset_06/fibsem_medulla_7col/README.md: -------------------------------------------------------------------------------- 1 | # FlyEM train/test dataset (FIBSEM, fly medulla 7 column dataset) 2 | --Srini Turaga :: June 3rd, 2015 3 | 4 | Here’s a very large set of training data assembled by the FlyEM project team 5 | here at Janelia. This is extracted from a part of the fly visual system known 6 | as the “medulla”, which is basically the second stage of computation starting 7 | at the photo-receptors -> lamina -> medulla. 8 | 9 | There are two 250^3 volumes (which they refer to as training volumes): 10 | 11 | /nobackup/turaga/data/fibsem_medulla_7col/trvol-250-1-h5/ 12 | /nobackup/turaga/data/fibsem_medulla_7col/trvol-250-2-h5/ 13 | 14 | And there are two 520^3 volumes (which they refer to as test volumes): 15 | 16 | /nobackup/turaga/data/fibsem_medulla_7col/tstvol-520-1-h5 17 | /nobackup/turaga/data/fibsem_medulla_7col/tstvol-520-2-h5 18 | 19 | Each of these volumes contains several files, but the most important ones are: 20 | 21 | img_normalized.h5: which contains the grayscale EM images in hdf5 format 22 | groundtruth_seg.h5: which contains the ground truth segmentation 23 | groundtruth_aff.h5: which contains the ground truth affinity graph 24 | 25 | 26 | ### History 27 | 28 | This dataset was assembled by reformatting a combination of files in the 29 | directories below into a consistent file format / naming convention. They were 30 | originally assembled by Toufiq, and I think the '-dawmr' directories were 31 | populated by Gary Huang / Viren Jain. 32 | 33 | /groups/flyem/data/viren_toufiq_comparison/trvol-250-1 34 | /groups/flyem/data/viren_toufiq_comparison/trvol-250-1-dawmr 35 | /groups/flyem/data/viren_toufiq_comparison/trvol-250-2 36 | /groups/flyem/data/viren_toufiq_comparison/trvol-250-2-dawmr 37 | /groups/flyem/data/viren_toufiq_comparison/tstvol-520-1 38 | /groups/flyem/data/viren_toufiq_comparison/tstvol-520-1-dawmr 39 | /groups/flyem/data/viren_toufiq_comparison/tstvol-520-2 40 | /groups/flyem/data/viren_toufiq_comparison/tstvol-520-2-dawmr 41 | 42 | 43 | -------------------------------------------------------------------------------- /dataset_06/malis/.gitignore: -------------------------------------------------------------------------------- 1 | build 2 | malis.cpp 3 | malis.so 4 | -------------------------------------------------------------------------------- /dataset_06/malis/README.md: -------------------------------------------------------------------------------- 1 | # MALIS 2 | #### Structured loss function for supervised learning of segmentation and clustering 3 | 4 | Python and MATLAB wrapper for C++ functions for computing the MALIS loss 5 | 6 | The MALIS loss is described here: 7 | 8 | SC Turaga, KL Briggman, M Helmstaedter, W Denk, HS Seung (2009). *Maximin learning of image segmentation*. _Advances in Neural Information Processing Systems (NIPS) 2009_. 9 | 10 | http://papers.nips.cc/paper/3887-maximin-affinity-learning-of-image-segmentation 11 | 12 | -------------------------------------------------------------------------------- /dataset_06/malis/__init__.py: -------------------------------------------------------------------------------- 1 | from malis import * 2 | -------------------------------------------------------------------------------- /dataset_06/malis/__init__.pyc: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/naibaf7/caffe_neural_models/9d372c4bc599029902185e19f89e5c39f842fff7/dataset_06/malis/__init__.pyc -------------------------------------------------------------------------------- /dataset_06/malis/make.sh: -------------------------------------------------------------------------------- 1 | python setup.py build_ext --inplace 2 | -------------------------------------------------------------------------------- /dataset_06/malis/malis_cpp.h: -------------------------------------------------------------------------------- 1 | #ifndef MALIS_CPP_H 2 | #define MALIS_CPP_H 3 | 4 | void connected_components_cpp(const int nVert, 5 | const int nEdge, const int* node1, const int* node2, const int* edgeWeight, 6 | int* seg); 7 | 8 | void malis_loss_weights_cpp(const int nVert, const int* seg, 9 | const int nEdge, const int* node1, const int* node2, const float* edgeWeight, 10 | const int pos, 11 | int* nPairPerEdge); 12 | 13 | void marker_watershed_cpp(const int nVert, const int* marker, 14 | const int nEdge, const int* node1, const int* node2, const float* edgeWeight, 15 | int* seg); 16 | #endif -------------------------------------------------------------------------------- /dataset_06/malis/setup.py: -------------------------------------------------------------------------------- 1 | from distutils.core import setup 2 | from distutils.extension import Extension 3 | from Cython.Distutils import build_ext 4 | 5 | ext_modules = [Extension("malis", ["malis.pyx", "malis_cpp.cpp"], language='c++',)] 6 | 7 | setup(cmdclass = {'build_ext': build_ext}, ext_modules = ext_modules) -------------------------------------------------------------------------------- /dataset_06/malis/test_malis.py: -------------------------------------------------------------------------------- 1 | import numpy as np 2 | import malis as m 3 | import h5py 4 | np.set_printoptions(precision=4) 5 | 6 | print "Can we make the `nhood' for an isotropic 3d dataset" 7 | print "corresponding to a 6-connected neighborhood?" 8 | nhood = m.mknhood3d(1) 9 | print nhood 10 | 11 | print "Can we make the `nhood' for an anisotropic 3d dataset" 12 | print "corresponding to a 4-connected neighborhood in-plane" 13 | print "and 26-connected neighborhood in the previous z-plane?" 14 | nhood2 = m.mknhood3d_aniso(1,1.8) 15 | print nhood2 16 | 17 | segTrue = np.array([0, 1, 1, 1, 2, 2, 0, 5, 5, 5, 5],dtype=np.int32); 18 | node1 = np.arange(segTrue.shape[0]-1,dtype=np.int32) 19 | node2 = np.arange(1,segTrue.shape[0],dtype=np.int32) 20 | nVert = segTrue.shape[0] 21 | edgeWeight = np.array([0, 1, 2, 0, 2, 0, 0, 1, 2, 2.5],dtype=np.float32); 22 | edgeWeight = edgeWeight/edgeWeight.max() 23 | print segTrue 24 | print edgeWeight 25 | 26 | nPairPos = m.malis_loss_weights(segTrue, node1, node2, edgeWeight, 1) 27 | nPairNeg = m.malis_loss_weights(segTrue, node1, node2, edgeWeight, 0) 28 | print np.vstack((nPairPos,nPairNeg)) 29 | # print nPairNeg 30 | 31 | idxkeep = (edgeWeight > 0).astype(np.int32) 32 | cc = m.connected_components(nVert,node1,node2,idxkeep) 33 | print cc 34 | 35 | 36 | # node1, node2 = m.nodelist_like((2,3,4),-np.eye(3)) 37 | # print node1 38 | # print node2 39 | 40 | datadir = '/groups/turaga/turagalab/greentea/project_data/dataset_06/fibsem_medulla_7col/trvol-250-1-h5/' 41 | print "Reading test volume from " + datadir 42 | # hdf5_raw_file = datadir + 'img_normalized.h5' 43 | hdf5_gt_file = datadir + 'groundtruth_seg.h5' 44 | # hdf5_aff_file = datadir + 'groundtruth_aff.h5' 45 | 46 | #hdf5_raw_file = 'zebrafish_friedrich/raw.hdf5' 47 | #hdf5_gt_file = 'zebrafish_friedrich/labels_2.hdf5' 48 | 49 | 50 | # hdf5_raw = h5py.File(hdf5_raw_file, 'r') 51 | h5seg = h5py.File(hdf5_gt_file, 'r') 52 | # hdf5_aff = h5py.File(hdf5_aff_file, 'r') 53 | 54 | seg = np.asarray(h5seg['main']).astype(np.int32) 55 | aff = m.seg_to_affgraph(seg,nhood) 56 | cc,ccSizes = m.connected_components_affgraph(aff,nhood) 57 | aff2 = m.seg_to_affgraph(cc,nhood) 58 | cc2,ccSizes2 = m.connected_components_affgraph(aff2,nhood) 59 | 60 | print "Comparing 'seg' and 'cc':" 61 | frac_disagree = np.mean(seg.ravel()!=cc.ravel()) 62 | ri,fscore,prec,rec = m.rand_index(seg,cc) 63 | V_rand,V_rand_split,V_rand_merge = m.compute_V_rand_N2(seg,cc) 64 | print "Connected components disagree at %f%% locations" % (frac_disagree*100) 65 | print "\tRand index: %f, fscore: %f, prec: %f, rec: %f" % (ri,fscore,prec,rec) 66 | print "\tV_rand: %f, V_rand_split: %f, V_rand_merge: %f" % (V_rand,V_rand_split,V_rand_merge) 67 | 68 | print "Comparing 'cc' and 'cc2':" 69 | frac_disagree = np.mean(cc.ravel()!=cc2.ravel()) 70 | ri,fscore,prec,rec = m.rand_index(cc,cc2) 71 | V_rand,V_rand_split,V_rand_merge = m.compute_V_rand_N2(cc,cc2) 72 | print "Connected components disagree at %f%% locations" % (frac_disagree*100) 73 | print "\tRand index: %f, fscore: %f, prec: %f, rec: %f" % (ri,fscore,prec,rec) 74 | print "\tV_rand: %f, V_rand_split: %f, V_rand_merge: %f" % (V_rand,V_rand_split,V_rand_merge) 75 | -------------------------------------------------------------------------------- /dataset_06/net/solver.prototxt: -------------------------------------------------------------------------------- 1 | train_net: "net/net_train.prototxt" 2 | base_lr: 0.00001 3 | momentum: 0.99 4 | weight_decay: 0.000005 5 | lr_policy: "inv" 6 | gamma: 0.0001 7 | power: 0.75 8 | max_iter: 100000 9 | snapshot: 2000 10 | snapshot_prefix: "net_" 11 | display: 50 12 | -------------------------------------------------------------------------------- /dataset_06/netconf.py: -------------------------------------------------------------------------------- 1 | import math 2 | 3 | # 10 GB total memory limit 4 | mem_global_limit = 10 * 1024 * 1024 * 1024 5 | 6 | # 4 GB single buffer memory limit 7 | mem_buf_limit = 4 * 1024 * 1024 * 1024 8 | 9 | # Desired number of output dimensions 10 | output_shape = [44, 44, 44] 11 | 12 | # Number of U-Net Pooling-Convolution downsampling/upsampling steps 13 | unet_depth = 3 14 | # Feature map increase rule (downsampling) 15 | def unet_fmap_inc_rule(fmaps): 16 | return int(math.ceil(fmaps * 3)); 17 | # Feature map decrease rule (upsampling) 18 | def unet_fmap_dec_rule(fmaps): 19 | return int(math.ceil(fmaps / 3)); 20 | 21 | # Skewed U-Net downsampling strategy 22 | unet_downsampling_strategy = [[1,2,2],[1,2,2],[1,2,2]] 23 | 24 | 25 | # Number of SK-Net Pooling-Convolution steps 26 | sknet_conv_depth = 0 27 | # Feature map increase rule 28 | def sknet_fmap_inc_rule(fmaps): 29 | return int(math.ceil(fmaps * 1.5)); 30 | # Number of 1x1 (IP) Convolution steps 31 | sknet_ip_depth = 0 32 | # Feature map increase rule from SK-Convolution to IP 33 | def sknet_fmap_bridge_rule(fmaps): 34 | return int(math.ceil(fmaps * 4)); 35 | # Feature map decrease rule within IP 36 | def sknet_fmap_dec_rule(fmaps): 37 | return int(math.ceil(fmaps / 2.5)); 38 | 39 | 40 | # Loss function and mode 41 | #loss_function: "malis", "euclid", "softmax" 42 | loss_function = "malis" 43 | -------------------------------------------------------------------------------- /dataset_06/netconf.pyc: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/naibaf7/caffe_neural_models/9d372c4bc599029902185e19f89e5c39f842fff7/dataset_06/netconf.pyc -------------------------------------------------------------------------------- /dataset_06/volume_slicer.pyc: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/naibaf7/caffe_neural_models/9d372c4bc599029902185e19f89e5c39f842fff7/dataset_06/volume_slicer.pyc -------------------------------------------------------------------------------- /dataset_08/.gitignore: -------------------------------------------------------------------------------- 1 | train 2 | validate_output 3 | validate_raw 4 | -------------------------------------------------------------------------------- /dataset_08/config.py: -------------------------------------------------------------------------------- 1 | # Specify the path caffe here 2 | caffe_path = "../../caffe_gt" 3 | # Specify wether or not to compile caffe 4 | library_compile = True 5 | 6 | # Specify the device to use 7 | device_id = 2 8 | 9 | # Specify the solver file 10 | solver_proto = "net/solver.prototxt" 11 | 12 | # Specify values for testing 13 | test_net = "net/net_test.prototxt" 14 | trained_model = "net_iter_12000.caffemodel" 15 | 16 | output_folder = "processed" 17 | 18 | output_dims = [44, 44, 44] 19 | input_padding = [388, 388, 388] 20 | 21 | border_reflect = False 22 | 23 | # Select "train" or "process" 24 | mode = "train" 25 | -------------------------------------------------------------------------------- /dataset_08/config.pyc: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/naibaf7/caffe_neural_models/9d372c4bc599029902185e19f89e5c39f842fff7/dataset_08/config.pyc -------------------------------------------------------------------------------- /dataset_08/device_query.sh: -------------------------------------------------------------------------------- 1 | ./../../caffe_neural_tool/build/caffe_neural_tool --devices 2 | -------------------------------------------------------------------------------- /dataset_08/net/solver.prototxt: -------------------------------------------------------------------------------- 1 | train_net: "net/net_train_malis.prototxt" 2 | base_lr: 0.0005 3 | momentum: 0.99 4 | weight_decay: 0.000005 5 | lr_policy: "inv" 6 | gamma: 0.0001 7 | power: 0.75 8 | max_iter: 100000 9 | snapshot: 2000 10 | snapshot_prefix: "net" 11 | display: 50 12 | -------------------------------------------------------------------------------- /dataset_08/netconf.py: -------------------------------------------------------------------------------- 1 | import math 2 | 3 | # 10 GB total memory limit 4 | mem_global_limit = 10 * 1024 * 1024 * 1024 5 | 6 | # 4 GB single buffer memory limit 7 | mem_buf_limit = 4 * 1024 * 1024 * 1024 8 | 9 | # Desired number of output dimensions 10 | output_shape = [44, 44] 11 | 12 | # Number of U-Net Pooling-Convolution downsampling/upsampling steps 13 | unet_depth = 4 14 | # Feature map increase rule (downsampling) 15 | def unet_fmap_inc_rule(fmaps): 16 | return int(math.ceil(fmaps * 2)); 17 | # Feature map decrease rule (upsampling) 18 | def unet_fmap_dec_rule(fmaps): 19 | return int(math.ceil(fmaps / 2)); 20 | 21 | # Skewed U-Net downsampling strategy 22 | unet_downsampling_strategy = [[2,2],[2,2],[2,2],[2,2]] 23 | 24 | 25 | # Number of SK-Net Pooling-Convolution steps 26 | sknet_conv_depth = 0 27 | # Feature map increase rule 28 | def sknet_fmap_inc_rule(fmaps): 29 | return int(math.ceil(fmaps * 1.5)); 30 | # Number of 1x1 (IP) Convolution steps 31 | sknet_ip_depth = 0 32 | # Feature map increase rule from SK-Convolution to IP 33 | def sknet_fmap_bridge_rule(fmaps): 34 | return int(math.ceil(fmaps * 4)); 35 | # Feature map decrease rule within IP 36 | def sknet_fmap_dec_rule(fmaps): 37 | return int(math.ceil(fmaps / 2.5)); 38 | 39 | 40 | # Loss function and mode 41 | #loss_function: "malis", "euclid", "softmax" 42 | loss_function = "softmax" 43 | -------------------------------------------------------------------------------- /dataset_08/netconf.pyc: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/naibaf7/caffe_neural_models/9d372c4bc599029902185e19f89e5c39f842fff7/dataset_08/netconf.pyc -------------------------------------------------------------------------------- /dataset_08/process.sh: -------------------------------------------------------------------------------- 1 | ./../../caffe_neural_tool/build/caffe_neural_tool --gpu 3 --graphic --process 0 --proto 'train_process.prototxt' 2 | -------------------------------------------------------------------------------- /dataset_08/train.sh: -------------------------------------------------------------------------------- 1 | ./../../caffe_neural_tool/build/caffe_neural_tool --gpu 3 --train 0 --proto 'train_process.prototxt' 2 | -------------------------------------------------------------------------------- /dataset_08/train_process.prototxt: -------------------------------------------------------------------------------- 1 | train { 2 | solverstate: "net_iter_22000.solverstate" 3 | solver: "net/solver.prototxt" 4 | input { 5 | padding_size: 184 6 | patch_size: 388 7 | channels: 3 8 | labels: 2 9 | batch_size: 1 10 | raw_images: "train/raw" 11 | label_images: "train/labels" 12 | preprocessor { 13 | normalization: true 14 | rotation: true 15 | mirror: true 16 | clahe { 17 | clip: 4.0 18 | } 19 | crop { 20 | imagecrop: 0 21 | labelcrop: 0 22 | } 23 | blur { 24 | mean: 0.0 25 | std: 0.1 26 | ksize: 5 27 | } 28 | histeq { 29 | patch_prior: false 30 | masking: false 31 | } 32 | } 33 | } 34 | filter_output { 35 | output_filters: false 36 | output: "u_filters" 37 | } 38 | } 39 | 40 | process { 41 | process_net: "net/net_test.prototxt" 42 | caffemodel: "net_iter_22000.caffemodel" 43 | input { 44 | padding_size: 184 45 | patch_size: 388 46 | channels: 3 47 | labels: 2 48 | batch_size: 1 49 | raw_images: "validate_raw" 50 | preprocessor { 51 | normalization: true 52 | clahe { 53 | clip: 4.0 54 | } 55 | crop { 56 | imagecrop: 0 57 | labelcrop: 0 58 | } 59 | } 60 | } 61 | filter_output { 62 | output_filters: false 63 | output: "u_filters" 64 | } 65 | output { 66 | format: "tif" 67 | fp32_out: true 68 | output: "validate_output" 69 | } 70 | } 71 | -------------------------------------------------------------------------------- /malis_setup/draw_net.sh: -------------------------------------------------------------------------------- 1 | BASEDIR=`pwd` 2 | (cd ../../caffe_gt/python/ && ./draw_net.py $BASEDIR/neuraltissue_net.prototxt $BASEDIR/neuraltissue_net.ps --rankdir 'TB' --margin '0, 0') 3 | ps2pdf -g5890x6820 neuraltissue_net.ps neuraltissue_net.pdf 4 | 5 | -------------------------------------------------------------------------------- /malis_setup/neuraltissue_net.prototxt: -------------------------------------------------------------------------------- 1 | name: "Neuraltissue-train" 2 | layer { 3 | include: {phase: TRAIN} 4 | name: "label" 5 | type: "MemoryData" 6 | top: "label" 7 | top: "labeli" 8 | memory_data_param { 9 | batch_size: 1 10 | channels: 1 11 | height: 128 12 | width: 128 13 | } 14 | } 15 | layer { 16 | include: {phase: TRAIN} 17 | name: "data" 18 | type: "MemoryData" 19 | top: "data" 20 | top: "datai" 21 | memory_data_param { 22 | batch_size: 1 23 | channels: 3 24 | height: 229 25 | width: 229 26 | } 27 | } 28 | layer { 29 | include: {phase: TRAIN} 30 | name: "silence" 31 | type: "Silence" 32 | bottom: "labeli" 33 | bottom: "datai" 34 | } 35 | 36 | layer { 37 | name: "ip" 38 | type: "Convolution" 39 | bottom: "data" 40 | top: "ip" 41 | param { 42 | lr_mult: 1 43 | } 44 | param { 45 | lr_mult: 2 46 | } 47 | convolution_param { 48 | num_output: 2 49 | kernel_size: 1 50 | stride: 1 51 | dilation: 1 52 | weight_filler { 53 | type: "gaussian" 54 | std: 0.01 55 | } 56 | bias_filler { 57 | type: "constant" 58 | } 59 | } 60 | } 61 | layer { 62 | name: "prob" 63 | type: "Softmax" 64 | bottom: "ip" 65 | top: "prob" 66 | } 67 | layer { 68 | include: {phase: TRAIN} 69 | name: "split" 70 | type: "Split" 71 | bottom: "label" 72 | top: "label_a" 73 | top: "label_b" 74 | } 75 | layer { 76 | include: {phase: TRAIN} 77 | name: "affinity" 78 | type: "Affinity" 79 | bottom: "prob" 80 | bottom: "label_a" 81 | top: "prob_affinity" 82 | top: "label_affinity" 83 | affinity_param { 84 | offset: 1 85 | offset: 0 86 | } 87 | } 88 | layer { 89 | include: {phase: TRAIN} 90 | name: "components" 91 | type: "ConnectedComponent" 92 | bottom: "label_b" 93 | top: "component" 94 | } 95 | layer { 96 | include: {phase: TRAIN} 97 | name: "loss" 98 | type: "MalisLoss" 99 | bottom: "prob_affinity" 100 | bottom: "label_affinity" 101 | bottom: "component" 102 | loss_param { 103 | } 104 | } 105 | -------------------------------------------------------------------------------- /net_old/neuraltissue_solver.prototxt: -------------------------------------------------------------------------------- 1 | # The training protocol buffer definition 2 | train_net: "neuraltissue_train.prototxt" 3 | ######################################################################## 4 | # The testing protocol buffer definition 5 | test_net: "neuraltissue_test.prototxt" 6 | ######################################################################## 7 | # Test_iter specifies how many forward passes the test should carry out. 8 | # it is the number of batches shown, then 9 | # examples shown = 'test_iter'*batch_size 10 | # Carry out testing every 'test_interval' training iterations. 11 | test_iter: 1000 12 | test_interval: 500 13 | ######################################################################## 14 | # The base learning rate, momentum and the weight decay of the network. 15 | # base_lr: 0.05 16 | base_lr: 0.01 17 | momentum: 0.9 18 | weight_decay: 0.0005 19 | ######################################################################## 20 | # The learning rate policy 21 | lr_policy: "inv" 22 | gamma: 0.0001 23 | power: 0.75 24 | #lr_policy: "step" 25 | #gamma: 0.1 26 | #stepsize: 20000 27 | ######################################################################## 28 | # The maximum number of iterations 29 | max_iter: 100000 30 | ######################################################################## 31 | # Snapshot intermediate results 32 | snapshot: 500 33 | snapshot_prefix: "neuraltissue" 34 | ######################################################################## 35 | # Display every 'display' iterations 36 | display: 50 37 | ######################################################################## 38 | # Solver mode: 0 for CPU and 1 for GPU 39 | solver_mode: 1 40 | ######################################################################## 41 | -------------------------------------------------------------------------------- /net_sk_2out/draw_net.sh: -------------------------------------------------------------------------------- 1 | BASEDIR=`pwd` 2 | (cd ../../caffe_gt/python/ && ./draw_net.py $BASEDIR/neuraltissue_net.prototxt $BASEDIR/neuraltissue_net.ps --rankdir 'TB' --margin '0, 0' --page '5, 8' --pagesize '5, 8' --size '5, 999') 3 | ps2pdf -g3600x5760 neuraltissue_net.ps neuraltissue_net.pdf 4 | 5 | -------------------------------------------------------------------------------- /net_sk_2out/neuraltissue_solver.prototxt: -------------------------------------------------------------------------------- 1 | # The training protocol buffer definition 2 | train_net: "../net_sk_2out/neuraltissue_net.prototxt" 3 | ######################################################################## 4 | # The testing protocol buffer definition 5 | # test_net: "../net_sk_2out/neuraltissue_net.prototxt" 6 | ######################################################################## 7 | # Test_iter specifies how many forward passes the test should carry out. 8 | # it is the number of batches shown, then 9 | # examples shown = 'test_iter'*batch_size 10 | # Carry out testing every 'test_interval' training iterations. 11 | # test_iter: 1000 12 | # test_interval: 500 13 | ######################################################################## 14 | # The base learning rate, momentum and the weight decay of the network. 15 | # base_lr: 0.05 16 | base_lr: 0.001 17 | momentum: 0.99 18 | weight_decay: 0.0005 19 | ######################################################################## 20 | # The learning rate policy 21 | lr_policy: "inv" 22 | gamma: 0.0001 23 | power: 0.75 24 | #lr_policy: "step" 25 | #gamma: 0.1 26 | #stepsize: 20000 27 | ######################################################################## 28 | # The maximum number of iterations 29 | max_iter: 100000 30 | ######################################################################## 31 | # Snapshot intermediate results 32 | snapshot: 2000 33 | snapshot_prefix: "neuraltissue_sk_2out" 34 | ######################################################################## 35 | # Display every 'display' iterations 36 | display: 50 37 | ######################################################################## 38 | -------------------------------------------------------------------------------- /net_sk_9out/neuraltissue_solver.prototxt: -------------------------------------------------------------------------------- 1 | # The training protocol buffer definition 2 | train_net: "../net_sk_9out/neuraltissue_net.prototxt" 3 | ######################################################################## 4 | # The testing protocol buffer definition 5 | # test_net: "../net_sk_9out/neuraltissue_net.prototxt" 6 | ######################################################################## 7 | # Test_iter specifies how many forward passes the test should carry out. 8 | # it is the number of batches shown, then 9 | # examples shown = 'test_iter'*batch_size 10 | # Carry out testing every 'test_interval' training iterations. 11 | # test_iter: 1000 12 | # test_interval: 500 13 | ######################################################################## 14 | # The base learning rate, momentum and the weight decay of the network. 15 | # base_lr: 0.05 16 | base_lr: 0.001 17 | momentum: 0.9 18 | weight_decay: 0.0005 19 | ######################################################################## 20 | # The learning rate policy 21 | lr_policy: "inv" 22 | gamma: 0.0001 23 | power: 0.75 24 | #lr_policy: "step" 25 | #gamma: 0.1 26 | #stepsize: 20000 27 | ######################################################################## 28 | # The maximum number of iterations 29 | max_iter: 100000 30 | ######################################################################## 31 | # Snapshot intermediate results 32 | snapshot: 2000 33 | snapshot_prefix: "neuraltissue_sk_9out" 34 | ######################################################################## 35 | # Display every 'display' iterations 36 | display: 50 37 | ######################################################################## 38 | -------------------------------------------------------------------------------- /net_u_2out/draw_net.sh: -------------------------------------------------------------------------------- 1 | BASEDIR=`pwd` 2 | (cd ../../caffe_gt/python/ && ./draw_net.py $BASEDIR/neuraltissue_net.prototxt $BASEDIR/neuraltissue_net.ps --rankdir 'LT' --margin '0, 0' --page '5, 8' --pagesize '5, 8' --size '5, 999') 3 | ps2pdf -g3600x5760 neuraltissue_net.ps neuraltissue_net.pdf 4 | 5 | -------------------------------------------------------------------------------- /net_u_2out/neuraltissue_solver.prototxt: -------------------------------------------------------------------------------- 1 | # The training protocol buffer definition 2 | train_net: "../net_u_2out/neuraltissue_net.prototxt" 3 | ######################################################################## 4 | # The testing protocol buffer definition 5 | # test_net: "../net_u_2out/neuraltissue_net.prototxt" 6 | ######################################################################## 7 | # Test_iter specifies how many forward passes the test should carry out. 8 | # it is the number of batches shown, then 9 | # examples shown = 'test_iter'*batch_size 10 | # Carry out testing every 'test_interval' training iterations. 11 | # test_iter: 1000 12 | # test_interval: 500 13 | ######################################################################## 14 | # The base learning rate, momentum and the weight decay of the network. 15 | # base_lr: 0.05 16 | base_lr: 0.00005 17 | momentum: 0.99 18 | weight_decay: 0.000005 19 | ######################################################################## 20 | # The learning rate policy 21 | lr_policy: "inv" 22 | gamma: 0.0001 23 | power: 0.75 24 | #lr_policy: "step" 25 | #gamma: 0.1 26 | #stepsize: 20000 27 | ######################################################################## 28 | # The maximum number of iterations 29 | max_iter: 100000 30 | ######################################################################## 31 | # Snapshot intermediate results 32 | snapshot: 2000 33 | snapshot_prefix: "neuraltissue_u_2out" 34 | ######################################################################## 35 | # Display every 'display' iterations 36 | display: 50 37 | ######################################################################## 38 | -------------------------------------------------------------------------------- /net_u_9out/neuraltissue_solver.prototxt: -------------------------------------------------------------------------------- 1 | # The training protocol buffer definition 2 | train_net: "../net_u_9out/neuraltissue_net.prototxt" 3 | ######################################################################## 4 | # The testing protocol buffer definition 5 | # test_net: "../net_u_9out/neuraltissue_net.prototxt" 6 | ######################################################################## 7 | # Test_iter specifies how many forward passes the test should carry out. 8 | # it is the number of batches shown, then 9 | # examples shown = 'test_iter'*batch_size 10 | # Carry out testing every 'test_interval' training iterations. 11 | # test_iter: 1000 12 | # test_interval: 500 13 | ######################################################################## 14 | # The base learning rate, momentum and the weight decay of the network. 15 | # base_lr: 0.05 16 | base_lr: 0.0001 17 | momentum: 0.99 18 | weight_decay: 0.00005 19 | ######################################################################## 20 | # The learning rate policy 21 | lr_policy: "inv" 22 | gamma: 0.0001 23 | power: 0.75 24 | #lr_policy: "step" 25 | #gamma: 0.1 26 | #stepsize: 20000 27 | ######################################################################## 28 | # The maximum number of iterations 29 | max_iter: 100000 30 | ######################################################################## 31 | # Snapshot intermediate results 32 | snapshot: 2000 33 | snapshot_prefix: "neuraltissue_u_9out" 34 | ######################################################################## 35 | # Display every 'display' iterations 36 | display: 50 37 | ######################################################################## 38 | -------------------------------------------------------------------------------- /net_u_small/neuraltissue_solver.prototxt: -------------------------------------------------------------------------------- 1 | # The training protocol buffer definition 2 | train_net: "../net_u_small/neuraltissue_net.prototxt" 3 | ######################################################################## 4 | # The testing protocol buffer definition 5 | # test_net: "../net_u_small/neuraltissue_net.prototxt" 6 | ######################################################################## 7 | # Test_iter specifies how many forward passes the test should carry out. 8 | # it is the number of batches shown, then 9 | # examples shown = 'test_iter'*batch_size 10 | # Carry out testing every 'test_interval' training iterations. 11 | # test_iter: 1000 12 | # test_interval: 500 13 | ######################################################################## 14 | # The base learning rate, momentum and the weight decay of the network. 15 | # base_lr: 0.05 16 | base_lr: 0.00005 17 | momentum: 0.99 18 | weight_decay: 0.000005 19 | ######################################################################## 20 | # The learning rate policy 21 | lr_policy: "inv" 22 | gamma: 0.0001 23 | power: 0.75 24 | #lr_policy: "step" 25 | #gamma: 0.1 26 | #stepsize: 20000 27 | ######################################################################## 28 | # The maximum number of iterations 29 | max_iter: 100000 30 | ######################################################################## 31 | # Snapshot intermediate results 32 | snapshot: 2000 33 | snapshot_prefix: "neuraltissue_u_2out" 34 | ######################################################################## 35 | # Display every 'display' iterations 36 | display: 50 37 | ######################################################################## 38 | -------------------------------------------------------------------------------- /net_u_small/script.py: -------------------------------------------------------------------------------- 1 | from __future__ import print_function 2 | import h5py 3 | import numpy as np 4 | from numpy import float32, int32, uint8, dtype 5 | 6 | # Relative path to where PyGreentea resides 7 | pygt_path = '../../PyGreentea' 8 | 9 | 10 | import sys, os 11 | sys.path.append(os.path.join(os.path.dirname(__file__), pygt_path)) 12 | 13 | # Other python modules 14 | import math 15 | 16 | # Load PyGreentea 17 | import PyGreentea as pygt 18 | 19 | # Create the network we want 20 | class NetConf: 21 | # 10 GB total memory limit 22 | mem_global_limit = 10 * 1024 * 1024 * 1024 23 | # 4 GB single buffer memory limit 24 | mem_buf_limit = 4 * 1024 * 1024 * 1024 25 | # Desired input dimensions (will select closest possible) 26 | input_shape = [380,380] 27 | # Desired output dimensions (will select closest posisble) 28 | output_shape = [196, 196] 29 | # Number of U-Net Pooling-Convolution downsampling/upsampling steps 30 | unet_depth = 4 31 | # Number of feature maps in the start 32 | fmap_start = 64 33 | # Number of input feature maps 34 | fmap_input = 3 35 | # Number of ouput feature maps 36 | fmap_output = 2 37 | # Feature map increase rule (downsampling) 38 | def unet_fmap_inc_rule(self, fmaps): 39 | return int(math.ceil(fmaps * 2)); 40 | # Feature map decrease rule (upsampling) 41 | def unet_fmap_dec_rule(self, fmaps): 42 | return int(math.ceil(fmaps / 2)); 43 | # Skewed U-Net downsampling strategy 44 | unet_downsampling_strategy = [[2,2],[2,2],[2,2],[2,2]] 45 | # Number of SK-Net Pooling-Convolution steps 46 | sknet_conv_depth = 0 47 | # Feature map increase rule 48 | def sknet_fmap_inc_rule(self, fmaps): 49 | return int(math.ceil(fmaps * 1.5)); 50 | # Number of 1x1 (IP) Convolution steps 51 | sknet_ip_depth = 0 52 | # Feature map increase rule from SK-Convolution to IP 53 | def sknet_fmap_bridge_rule(self, fmaps): 54 | return int(math.ceil(fmaps * 4)); 55 | # Feature map decrease rule within IP 56 | def sknet_fmap_dec_rule(self, fmaps): 57 | return int(math.ceil(fmaps / 2.5)); 58 | # Loss function and mode ("malis", "euclid", "softmax") 59 | loss_function = "softmax" 60 | 61 | netconf = NetConf() 62 | 63 | train_net_conf, test_net_conf = pygt.netgen.create_nets(netconf) 64 | 65 | 66 | with open('net_train.prototxt', 'w') as f: 67 | print(train_net_conf, file=f) 68 | 69 | with open('net_test.prototxt', 'w') as f: 70 | print(test_net_conf, file=f) 71 | 72 | -------------------------------------------------------------------------------- /net_upsample/neuraltissue_net.prototxt: -------------------------------------------------------------------------------- 1 | name: "Neuraltissue-process" 2 | layer { 3 | name: "data" 4 | type: "MemoryData" 5 | top: "data" 6 | top: "label" 7 | memory_data_param { 8 | batch_size: 1 9 | channels: 3 10 | height: 256 11 | width: 256 12 | } 13 | } 14 | layer { 15 | name: "silence" 16 | type: "Silence" 17 | bottom: "label" 18 | } 19 | layer { 20 | name: "upconv" 21 | type: "Deconvolution" 22 | bottom: "data" 23 | top: "upconv" 24 | param { 25 | lr_mult: 0 26 | decay_mult: 0 27 | } 28 | convolution_param { 29 | num_output: 3 30 | group: 3 31 | kernel_size: 2 32 | stride: 2 33 | weight_filler { 34 | type: "constant" 35 | value: 1 36 | } 37 | bias_term: false 38 | } 39 | } 40 | -------------------------------------------------------------------------------- /net_usk_2out/draw_net.sh: -------------------------------------------------------------------------------- 1 | BASEDIR=`pwd` 2 | (cd ../../caffe_gt/python/ && ./draw_net.py $BASEDIR/neuraltissue_net.prototxt $BASEDIR/neuraltissue_net.ps --rankdir 'LT' --margin '0, 0' --page '5, 8' --pagesize '5, 8' --size '5, 999') 3 | ps2pdf -g3600x5760 neuraltissue_net.ps neuraltissue_net.pdf 4 | 5 | -------------------------------------------------------------------------------- /net_usk_2out/neuraltissue_solver.prototxt: -------------------------------------------------------------------------------- 1 | # The training protocol buffer definition 2 | train_net: "../net_usk_2out/neuraltissue_net.prototxt" 3 | ######################################################################## 4 | # The testing protocol buffer definition 5 | # test_net: "../net_usk_2out/neuraltissue_net.prototxt" 6 | ######################################################################## 7 | # Test_iter specifies how many forward passes the test should carry out. 8 | # it is the number of batches shown, then 9 | # examples shown = 'test_iter'*batch_size 10 | # Carry out testing every 'test_interval' training iterations. 11 | # test_iter: 1000 12 | # test_interval: 500 13 | ######################################################################## 14 | # The base learning rate, momentum and the weight decay of the network. 15 | # base_lr: 0.05 16 | base_lr: 0.001 17 | momentum: 0.99 18 | weight_decay: 0.0005 19 | ######################################################################## 20 | # The learning rate policy 21 | lr_policy: "inv" 22 | gamma: 0.0001 23 | power: 0.75 24 | #lr_policy: "step" 25 | #gamma: 0.1 26 | #stepsize: 20000 27 | ######################################################################## 28 | # The maximum number of iterations 29 | max_iter: 100000 30 | ######################################################################## 31 | # Snapshot intermediate results 32 | snapshot: 2000 33 | snapshot_prefix: "neuraltissue_usk_2out" 34 | ######################################################################## 35 | # Display every 'display' iterations 36 | display: 50 37 | ######################################################################## 38 | -------------------------------------------------------------------------------- /net_usk_3out/neuraltissue_solver.prototxt: -------------------------------------------------------------------------------- 1 | # The training protocol buffer definition 2 | train_net: "../net_usk_3out/neuraltissue_net.prototxt" 3 | ######################################################################## 4 | # The testing protocol buffer definition 5 | # test_net: "../net_usk_3out/neuraltissue_net.prototxt" 6 | ######################################################################## 7 | # Test_iter specifies how many forward passes the test should carry out. 8 | # it is the number of batches shown, then 9 | # examples shown = 'test_iter'*batch_size 10 | # Carry out testing every 'test_interval' training iterations. 11 | # test_iter: 1000 12 | # test_interval: 500 13 | ######################################################################## 14 | # The base learning rate, momentum and the weight decay of the network. 15 | # base_lr: 0.05 16 | base_lr: 0.0005 17 | momentum: 0.99 18 | weight_decay: 0.00005 19 | ######################################################################## 20 | # The learning rate policy 21 | lr_policy: "inv" 22 | gamma: 0.0001 23 | power: 0.75 24 | #lr_policy: "step" 25 | #gamma: 0.1 26 | #stepsize: 20000 27 | ######################################################################## 28 | # The maximum number of iterations 29 | max_iter: 100000 30 | ######################################################################## 31 | # Snapshot intermediate results 32 | snapshot: 2000 33 | snapshot_prefix: "neuraltissue_usk_3out" 34 | ######################################################################## 35 | # Display every 'display' iterations 36 | display: 50 37 | ######################################################################## 38 | -------------------------------------------------------------------------------- /net_usk_9out/neuraltissue_solver.prototxt: -------------------------------------------------------------------------------- 1 | # The training protocol buffer definition 2 | train_net: "../net_usk_9out/neuraltissue_net.prototxt" 3 | ######################################################################## 4 | # The testing protocol buffer definition 5 | # test_net: "../net_usk_9out/neuraltissue_net.prototxt" 6 | ######################################################################## 7 | # Test_iter specifies how many forward passes the test should carry out. 8 | # it is the number of batches shown, then 9 | # examples shown = 'test_iter'*batch_size 10 | # Carry out testing every 'test_interval' training iterations. 11 | # test_iter: 1000 12 | # test_interval: 500 13 | ######################################################################## 14 | # The base learning rate, momentum and the weight decay of the network. 15 | # base_lr: 0.05 16 | base_lr: 0.0005 17 | momentum: 0.99 18 | weight_decay: 0.00005 19 | ######################################################################## 20 | # The learning rate policy 21 | lr_policy: "inv" 22 | gamma: 0.0001 23 | power: 0.75 24 | #lr_policy: "step" 25 | #gamma: 0.1 26 | #stepsize: 20000 27 | ######################################################################## 28 | # The maximum number of iterations 29 | max_iter: 100000 30 | ######################################################################## 31 | # Snapshot intermediate results 32 | snapshot: 2000 33 | snapshot_prefix: "neuraltissue_usk_9out" 34 | ######################################################################## 35 | # Display every 'display' iterations 36 | display: 50 37 | ######################################################################## 38 | -------------------------------------------------------------------------------- /pygt_01/loss.mat: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/naibaf7/caffe_neural_models/9d372c4bc599029902185e19f89e5c39f842fff7/pygt_01/loss.mat -------------------------------------------------------------------------------- /pygt_example/.gitignore: -------------------------------------------------------------------------------- 1 | *.prototxt 2 | -------------------------------------------------------------------------------- /pygt_example/draw_net.sh: -------------------------------------------------------------------------------- 1 | BASEDIR=`pwd` 2 | (cd ../../caffe_gt/python/ && ./draw_net.py $BASEDIR/net_train_malis.prototxt $BASEDIR/net.ps --rankdir 'TB' --margin '0, 0' --page '5, 8' --pagesize '5, 8' --size '5, 999') 3 | ps2pdf -g3600x5760 net.ps net.pdf 4 | 5 | -------------------------------------------------------------------------------- /pygt_fibsem/.gitignore: -------------------------------------------------------------------------------- 1 | *.h5 2 | -------------------------------------------------------------------------------- /pygt_fibsem/process.py: -------------------------------------------------------------------------------- 1 | from __future__ import print_function 2 | import h5py 3 | import numpy as np 4 | from numpy import float32, int32, uint8, dtype 5 | import sys 6 | 7 | # Relative path to where PyGreentea resides 8 | pygt_path = '../../PyGreentea' 9 | 10 | 11 | import sys, os 12 | sys.path.append(os.path.join(os.path.dirname(__file__), pygt_path)) 13 | 14 | # Other python modules 15 | import math 16 | 17 | # Load PyGreentea 18 | import PyGreentea as pygt 19 | 20 | 21 | 22 | test_net_file = 'net_test.prototxt' 23 | test_device = 3 24 | 25 | pygt.caffe.set_devices((test_device,)) 26 | 27 | caffemodels = pygt.getCaffeModels('net'); 28 | 29 | test_net = pygt.init_testnet(test_net_file, trained_model=caffemodels[-1][1], test_device=test_device) 30 | 31 | hdf5_raw_file = '../dataset_06/fibsem_medulla_7col/trvol-250-1-h5/img_normalized.h5' 32 | hdf5_raw = h5py.File(hdf5_raw_file, 'r') 33 | hdf5_raw_ds = pygt.normalize(np.asarray(hdf5_raw[hdf5_raw.keys()[0]]).astype(float32), -1, 1) 34 | test_dataset = {} 35 | test_dataset['data'] = hdf5_raw_ds 36 | 37 | pred_array = pygt.process(test_net, [test_dataset]) 38 | 39 | outhdf5 = h5py.File('test_out.h5', 'w') 40 | outdset = outhdf5.create_dataset('main', np.shape(pred_array)[1:], np.float32, data=pred_array) 41 | outdset.attrs['label'] = np.string_('-1,0,0;0,-1,0;0,0,-1') 42 | outhdf5.close() -------------------------------------------------------------------------------- /pygt_isbi2012/mknet.py: -------------------------------------------------------------------------------- 1 | from __future__ import print_function 2 | import h5py 3 | import numpy as np 4 | from numpy import float32, int32, uint8, dtype 5 | import sys 6 | 7 | # Relative path to where PyGreentea resides 8 | pygt_path = '../../PyGreentea' 9 | 10 | 11 | import sys, os 12 | sys.path.append(os.path.join(os.path.dirname(__file__), pygt_path)) 13 | 14 | # Other python modules 15 | import math 16 | 17 | # Load PyGreentea 18 | import PyGreentea as pygt 19 | 20 | 21 | netconf = pygt.netgen.NetConf() 22 | 23 | netconf.use_batchnorm = False 24 | netconf.dropout = 0.0 25 | 26 | netconf.fmap_start = 4 27 | 28 | #sk1 = pygt.netgen.SKNetConf() 29 | #sk1.sknet_conv = [[8],[6],[4]] 30 | 31 | #sk2 = pygt.netgen.SKNetConf() 32 | #sk2.sknet_padding = [44,44] 33 | #sk2.sknet_conv = [[6],[4],[2]] 34 | 35 | netconf.u_netconfs[0].unet_depth = 3 36 | netconf.u_netconfs[0].unet_fmap_inc_rule = lambda fmaps: int(math.ceil(float(fmaps) * 1)) 37 | netconf.u_netconfs[0].unet_fmap_dec_rule = lambda fmaps: int(math.ceil(float(fmaps) / 1)) 38 | # netconf.u_netconfs[0].sk_netconfs = [None,sk1,sk2] 39 | # netconf.u_netconfs = [netconf.u_netconfs[0],netconf.u_netconfs[0]] 40 | 41 | 42 | netconf.loss_function = "euclid" 43 | inshape,outshape,fmaps = pygt.netgen.compute_valid_io_shapes(netconf,pygt.netgen.caffe_pb2.TRAIN,[30,30],[200,200],constraints=[None,lambda x: x[0], lambda x: x[1]]) 44 | 45 | 46 | # We choose the maximum that still gives us 20 fmaps: 47 | index = [n for n, i in enumerate(fmaps) if i>=4][-1] 48 | print("Index to use: %s" % index) 49 | # Some patching to allow our new parameters 50 | netconf.input_shape = inshape[index] 51 | netconf.output_shape = outshape[index] 52 | netconf.fmap_start = 4 53 | 54 | netconf.loss_function = "euclid" 55 | train_net_conf_euclid, test_net_conf = pygt.netgen.create_nets(netconf) 56 | netconf.loss_function = "malis" 57 | train_net_conf_malis, test_net_conf = pygt.netgen.create_nets(netconf) 58 | 59 | with open('net_train_euclid.prototxt', 'w') as f: 60 | print(train_net_conf_euclid, file=f) 61 | with open('net_train_malis.prototxt', 'w') as f: 62 | print(train_net_conf_malis, file=f) 63 | with open('net_test.prototxt', 'w') as f: 64 | print(test_net_conf, file=f) 65 | 66 | -------------------------------------------------------------------------------- /pygt_isbi2012/process.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/naibaf7/caffe_neural_models/9d372c4bc599029902185e19f89e5c39f842fff7/pygt_isbi2012/process.py -------------------------------------------------------------------------------- /pygt_isbi2012/train.py: -------------------------------------------------------------------------------- 1 | from __future__ import print_function 2 | import h5py 3 | import numpy as np 4 | from numpy import float32, int32, uint8, dtype 5 | import sys 6 | 7 | # Relative path to where PyGreentea resides 8 | pygt_path = '../../PyGreentea' 9 | 10 | 11 | import sys, os 12 | sys.path.append(os.path.join(os.path.dirname(__file__), pygt_path)) 13 | 14 | # Other python modules 15 | import math 16 | 17 | # Load PyGreentea 18 | import PyGreentea as pygt 19 | 20 | # Load the datasets 21 | hdf5_raw_file = '../dataset_06/fibsem_medulla_7col/tstvol-520-1-h5/img_normalized.h5' 22 | hdf5_gt_file = '../dataset_06/fibsem_medulla_7col/tstvol-520-1-h5/groundtruth_seg.h5' 23 | hdf5_aff_file = '../dataset_06/fibsem_medulla_7col/tstvol-520-1-h5/groundtruth_aff.h5' 24 | 25 | hdf5_raw = h5py.File(hdf5_raw_file, 'r') 26 | hdf5_gt = h5py.File(hdf5_gt_file, 'r') 27 | hdf5_aff = h5py.File(hdf5_aff_file, 'r') 28 | 29 | hdf5_raw_ds = pygt.normalize(np.asarray(hdf5_raw[hdf5_raw.keys()[0]]).astype(float32), -1, 1) 30 | hdf5_gt_ds = np.asarray(hdf5_gt[hdf5_gt.keys()[0]]).astype(float32) 31 | hdf5_aff_ds = np.asarray(hdf5_aff[hdf5_aff.keys()[0]]).astype(float32) 32 | 33 | 34 | dataset = {} 35 | dataset['data'] = hdf5_raw_ds[None, :] 36 | dataset['components'] = hdf5_gt_ds[None, :] 37 | dataset['label'] = hdf5_aff_ds[:] 38 | dataset['nhood'] = pygt.malis.mknhood2d() 39 | 40 | #test_dataset = {} 41 | #test_dataset['data'] = hdf5_raw_ds 42 | #test_dataset['label'] = hdf5_aff_ds 43 | 44 | 45 | # Set train options 46 | class TrainOptions: 47 | loss_function = "euclid" 48 | loss_output_file = "log/loss.log" 49 | test_output_file = "log/test.log" 50 | test_interval = 4000 51 | scale_error = True 52 | training_method = "affinity" 53 | recompute_affinity = True 54 | train_device = 0 55 | test_device = 0 56 | test_net=None #'net_test.prototxt' 57 | 58 | 59 | options = TrainOptions() 60 | 61 | # Set solver options 62 | solver_config = pygt.caffe.SolverParameter() 63 | solver_config.train_net = 'net_train_euclid.prototxt' 64 | solver_config.base_lr = 0.00001 65 | solver_config.momentum = 0.99 66 | solver_config.weight_decay = 0.000005 67 | solver_config.lr_policy = 'inv' 68 | solver_config.gamma = 0.0001 69 | solver_config.power = 0.75 70 | solver_config.max_iter = 8000 71 | solver_config.snapshot = 2000 72 | solver_config.snapshot_prefix = 'net' 73 | solver_config.type = 'Adam' 74 | solver_config.display = 1 75 | 76 | # Set devices 77 | # pygt.caffe.enumerate_devices(False) 78 | pygt.caffe.set_devices((options.train_device,)) 79 | 80 | 81 | solverstates = pygt.getSolverStates(solver_config.snapshot_prefix); 82 | 83 | # First training method 84 | if (len(solverstates) == 0 or solverstates[-1][0] < solver_config.max_iter): 85 | solver, test_net = pygt.init_solver(solver_config, options) 86 | if (len(solverstates) > 0): 87 | solver.restore(solverstates[-1][1]) 88 | pygt.train(solver, test_net, [dataset], [], options) 89 | 90 | 91 | 92 | 93 | 94 | 95 | 96 | -------------------------------------------------------------------------------- /pygt_snemi_aniso/mknet.py: -------------------------------------------------------------------------------- 1 | from __future__ import print_function 2 | import h5py 3 | import numpy as np 4 | from numpy import float32, int32, uint8, dtype 5 | import sys 6 | 7 | # Relative path to where PyGreentea resides 8 | pygt_path = '../../PyGreentea' 9 | 10 | 11 | import sys, os 12 | sys.path.append(os.path.join(os.path.dirname(__file__), pygt_path)) 13 | 14 | # Other python modules 15 | import math 16 | 17 | # Load PyGreentea 18 | import PyGreentea as pygt 19 | 20 | # Create the network we want 21 | class NetConf: 22 | # 10 GB total memory limit 23 | mem_global_limit = 10 * 1024 * 1024 * 1024 24 | # 4 GB single buffer memory limit 25 | mem_buf_limit = 4 * 1024 * 1024 * 1024 26 | # Desired input dimensions (will select closest possible) 27 | input_shape = [44,132,132] 28 | # Desired output dimensions (will select closest posisble) 29 | output_shape = [16, 44, 44] 30 | # Number of U-Net Pooling-Convolution downsampling/upsampling steps 31 | unet_depth = 3 32 | # Number of feature maps in the start 33 | fmap_start = 16 34 | # Number of input feature maps 35 | fmap_input = 1 36 | # Number of ouput feature maps 37 | fmap_output = 11 38 | # Feature map increase rule (downsampling) 39 | def unet_fmap_inc_rule(self, fmaps): 40 | return int(math.ceil(fmaps * 3)); 41 | # Feature map decrease rule (upsampling) 42 | def unet_fmap_dec_rule(self, fmaps): 43 | return int(math.ceil(fmaps / 3)); 44 | # Skewed U-Net downsampling strategy 45 | unet_downsampling_strategy = [[1,2,2],[1,2,2],[1,2,2]] 46 | # Number of SK-Net Pooling-Convolution steps 47 | sknet_conv_depth = 0 48 | # Feature map increase rule 49 | def sknet_fmap_inc_rule(self, fmaps): 50 | return int(math.ceil(fmaps * 1.5)); 51 | # Number of 1x1 (IP) Convolution steps 52 | sknet_ip_depth = 0 53 | # Feature map increase rule from SK-Convolution to IP 54 | def sknet_fmap_bridge_rule(self, fmaps): 55 | return int(math.ceil(fmaps * 4)); 56 | # Feature map decrease rule within IP 57 | def sknet_fmap_dec_rule(self, fmaps): 58 | return int(math.ceil(fmaps / 2.5)); 59 | # Loss function and mode ("malis", "euclid", "softmax") 60 | loss_function = "euclid" 61 | 62 | 63 | netconf = NetConf() 64 | netconf.loss_function = "euclid" 65 | train_net_conf_euclid, test_net_conf = pygt.netgen.create_nets(netconf) 66 | netconf.loss_function = "malis" 67 | train_net_conf_malis, test_net_conf = pygt.netgen.create_nets(netconf) 68 | 69 | with open('net_train_euclid.prototxt', 'w') as f: 70 | print(train_net_conf_euclid, file=f) 71 | with open('net_train_malis.prototxt', 'w') as f: 72 | print(train_net_conf_malis, file=f) 73 | with open('net_test.prototxt', 'w') as f: 74 | print(test_net_conf, file=f) 75 | -------------------------------------------------------------------------------- /pygt_uvisual/.gitignore: -------------------------------------------------------------------------------- 1 | dump 2 | -------------------------------------------------------------------------------- /pygt_uvisual/mknet.py: -------------------------------------------------------------------------------- 1 | from __future__ import print_function 2 | import h5py 3 | import numpy as np 4 | from numpy import float32, int32, uint8, dtype 5 | import sys 6 | 7 | # Relative path to where PyGreentea resides 8 | pygt_path = '../../PyGreentea' 9 | 10 | 11 | import sys, os 12 | sys.path.append(os.path.join(os.path.dirname(__file__), pygt_path)) 13 | 14 | # Other python modules 15 | import math 16 | 17 | # Load PyGreentea 18 | import PyGreentea as pygt 19 | 20 | 21 | netconf = pygt.netgen.NetConf() 22 | 23 | netconf.use_batchnorm = False 24 | netconf.dropout = 0.0 25 | 26 | netconf.fmap_start = 4 27 | 28 | #sk1 = pygt.netgen.SKNetConf() 29 | #sk1.sknet_conv = [[8],[6],[4]] 30 | 31 | #sk2 = pygt.netgen.SKNetConf() 32 | #sk2.sknet_padding = [44,44] 33 | #sk2.sknet_conv = [[6],[4],[2]] 34 | 35 | netconf.u_netconfs[0].unet_depth = 3 36 | netconf.u_netconfs[0].unet_fmap_inc_rule = lambda fmaps: int(math.ceil(float(fmaps) * 1)) 37 | netconf.u_netconfs[0].unet_fmap_dec_rule = lambda fmaps: int(math.ceil(float(fmaps) / 1)) 38 | # netconf.u_netconfs[0].sk_netconfs = [None,sk1,sk2] 39 | # netconf.u_netconfs = [netconf.u_netconfs[0],netconf.u_netconfs[0]] 40 | 41 | 42 | netconf.loss_function = "euclid" 43 | inshape,outshape,fmaps = pygt.netgen.compute_valid_io_shapes(netconf,pygt.netgen.caffe_pb2.TRAIN,[30,30],[200,200],constraints=[None,lambda x: x[0], lambda x: x[1]]) 44 | 45 | 46 | # We choose the maximum that still gives us 20 fmaps: 47 | index = [n for n, i in enumerate(fmaps) if i>=4][-1] 48 | print("Index to use: %s" % index) 49 | # Some patching to allow our new parameters 50 | netconf.input_shape = inshape[index] 51 | netconf.output_shape = outshape[index] 52 | netconf.fmap_start = 4 53 | 54 | netconf.loss_function = "euclid" 55 | train_net_conf_euclid, test_net_conf, train_net_tikzgraph, test_net_tikzgraph = pygt.netgen.create_nets(netconf) 56 | netconf.loss_function = "malis" 57 | train_net_conf_malis, test_net_conf, train_net_tikzgraph, test_net_tikzgraph = pygt.netgen.create_nets(netconf) 58 | 59 | with open('net_train_euclid.prototxt', 'w') as f: 60 | print(train_net_conf_euclid, file=f) 61 | with open('net_train_malis.prototxt', 'w') as f: 62 | print(train_net_conf_malis, file=f) 63 | with open('net_test.prototxt', 'w') as f: 64 | print(test_net_conf, file=f) 65 | 66 | with open('trainnet.tex', 'w') as f: 67 | print(train_net_tikzgraph, file=f) 68 | with open('testnet.tex', 'w') as f: 69 | print(test_net_tikzgraph, file=f) 70 | 71 | -------------------------------------------------------------------------------- /pygt_uvisual/process.py: -------------------------------------------------------------------------------- 1 | from __future__ import print_function 2 | import h5py 3 | import numpy as np 4 | from numpy import float32, int32, uint8, dtype 5 | import sys 6 | 7 | # Relative path to where PyGreentea resides 8 | pygt_path = '../../PyGreentea' 9 | 10 | 11 | import sys, os 12 | sys.path.append(os.path.join(os.path.dirname(__file__), pygt_path)) 13 | 14 | # Other python modules 15 | import math 16 | 17 | # Load PyGreentea 18 | import PyGreentea as pygt 19 | 20 | 21 | 22 | test_net_file = 'net_test.prototxt' 23 | test_device = 0 24 | 25 | pygt.caffe.set_devices((test_device,)) 26 | 27 | caffemodels = pygt.getCaffeModels('net'); 28 | 29 | test_net = pygt.init_testnet(test_net_file, trained_model=caffemodels[-1][1], test_device=test_device) 30 | 31 | 32 | # Load the datasets 33 | hdf5_raw_file = '../dataset_06/fibsem_medulla_7col/tstvol-520-1-h5/img_normalized.h5' 34 | hdf5_gt_file = '../dataset_06/fibsem_medulla_7col/tstvol-520-1-h5/groundtruth_seg.h5' 35 | hdf5_aff_file = '../dataset_06/fibsem_medulla_7col/tstvol-520-1-h5/groundtruth_aff.h5' 36 | 37 | hdf5_raw = h5py.File(hdf5_raw_file, 'r') 38 | hdf5_gt = h5py.File(hdf5_gt_file, 'r') 39 | hdf5_aff = h5py.File(hdf5_aff_file, 'r') 40 | 41 | hdf5_raw_ds = pygt.normalize(np.asarray(hdf5_raw[hdf5_raw.keys()[0]]).astype(float32), -1, 1) 42 | hdf5_gt_ds = np.asarray(hdf5_gt[hdf5_gt.keys()[0]]).astype(float32) 43 | hdf5_aff_ds = np.asarray(hdf5_aff[hdf5_aff.keys()[0]]).astype(float32) 44 | 45 | datasets = [] 46 | for i in range(0,1): 47 | dataset = {} 48 | dataset['data'] = hdf5_raw_ds[None, i, :] 49 | datasets += [dataset] 50 | 51 | 52 | pred_array = pygt.process(test_net, datasets) 53 | 54 | pygt.dump_tikzgraph_maps(test_net, 'dump') 55 | 56 | outhdf5 = h5py.File('test_out.h5', 'w') 57 | outdset = outhdf5.create_dataset('main', np.shape(pred_array)[1:], np.float32, data=pred_array) 58 | outhdf5.close() -------------------------------------------------------------------------------- /pygt_uvisual/test_out.h5: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/naibaf7/caffe_neural_models/9d372c4bc599029902185e19f89e5c39f842fff7/pygt_uvisual/test_out.h5 -------------------------------------------------------------------------------- /pygt_uvisual/testnet.aux: -------------------------------------------------------------------------------- 1 | \relax 2 | \providecommand\hyper@newdestlabel[2]{} 3 | \providecommand\BKM@entry[2]{} 4 | \providecommand\HyperFirstAtBeginDocument{\AtBeginDocument} 5 | \HyperFirstAtBeginDocument{\ifx\hyper@anchor\@undefined 6 | \global\let\oldcontentsline\contentsline 7 | \gdef\contentsline#1#2#3#4{\oldcontentsline{#1}{#2}{#3}} 8 | \global\let\oldnewlabel\newlabel 9 | \gdef\newlabel#1#2{\newlabelxx{#1}#2} 10 | \gdef\newlabelxx#1#2#3#4#5#6{\oldnewlabel{#1}{{#2}{#3}}} 11 | \AtEndDocument{\ifx\hyper@anchor\@undefined 12 | \let\contentsline\oldcontentsline 13 | \let\newlabel\oldnewlabel 14 | \fi} 15 | \fi} 16 | \global\let\hyper@last\relax 17 | \gdef\HyperFirstAtBeginDocument#1{#1} 18 | \providecommand\HyField@AuxAddToFields[1]{} 19 | \providecommand\HyField@AuxAddToCoFields[2]{} 20 | \abx@aux@sortscheme{none} 21 | \providecommand \oddpage@label [2]{} 22 | \@writefile{toc}{\boolfalse {citerequest}\boolfalse {citetracker}\boolfalse {pagetracker}\boolfalse {backtracker}\relax } 23 | \@writefile{lof}{\boolfalse {citerequest}\boolfalse {citetracker}\boolfalse {pagetracker}\boolfalse {backtracker}\relax } 24 | \@writefile{lot}{\boolfalse {citerequest}\boolfalse {citetracker}\boolfalse {pagetracker}\boolfalse {backtracker}\relax } 25 | \select@language{british} 26 | \@writefile{toc}{\defcounter {refsection}{0}\relax }\@writefile{toc}{\select@language{british}} 27 | \@writefile{lof}{\defcounter {refsection}{0}\relax }\@writefile{lof}{\select@language{british}} 28 | \@writefile{lot}{\defcounter {refsection}{0}\relax }\@writefile{lot}{\select@language{british}} 29 | -------------------------------------------------------------------------------- /pygt_uvisual/testnet.run.xml: -------------------------------------------------------------------------------- 1 | 2 | 3 | 4 | 5 | 7 | 8 | 9 | 10 | 11 | 12 | 13 | 14 | 15 | 16 | 17 | 18 | 19 | 20 | 23 | 28 | 33 | 36 | 39 | 42 | ]> 43 | 44 | 45 | latex 46 | 47 | testnet.bcf 48 | 49 | 50 | testnet.bbl 51 | 52 | 53 | blx-dm.def 54 | blx-compat.def 55 | biblatex.def 56 | blx-natbib.def 57 | numeric.bbx 58 | standard.bbx 59 | numeric.cbx 60 | biblatex.cfg 61 | british.lbx 62 | english.lbx 63 | 64 | 65 | 66 | biber 67 | 68 | biber 69 | testnet 70 | 71 | 72 | testnet.bcf 73 | 74 | 75 | testnet.bbl 76 | 77 | 78 | testnet.bbl 79 | 80 | 81 | testnet.bcf 82 | 83 | 84 | 85 | -------------------------------------------------------------------------------- /pygt_uvisual/testnet.synctex.gz: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/naibaf7/caffe_neural_models/9d372c4bc599029902185e19f89e5c39f842fff7/pygt_uvisual/testnet.synctex.gz -------------------------------------------------------------------------------- /pygt_uvisual/train.py: -------------------------------------------------------------------------------- 1 | from __future__ import print_function 2 | import h5py 3 | import numpy as np 4 | from numpy import float32, int32, uint8, dtype 5 | import sys 6 | 7 | # Relative path to where PyGreentea resides 8 | pygt_path = '../../PyGreentea' 9 | 10 | 11 | import sys, os 12 | sys.path.append(os.path.join(os.path.dirname(__file__), pygt_path)) 13 | 14 | # Other python modules 15 | import math 16 | 17 | # Load PyGreentea 18 | import PyGreentea as pygt 19 | 20 | # Load the datasets 21 | hdf5_raw_file = '../dataset_06/fibsem_medulla_7col/tstvol-520-1-h5/img_normalized.h5' 22 | hdf5_gt_file = '../dataset_06/fibsem_medulla_7col/tstvol-520-1-h5/groundtruth_seg.h5' 23 | hdf5_aff_file = '../dataset_06/fibsem_medulla_7col/tstvol-520-1-h5/groundtruth_aff.h5' 24 | 25 | hdf5_raw = h5py.File(hdf5_raw_file, 'r') 26 | hdf5_gt = h5py.File(hdf5_gt_file, 'r') 27 | hdf5_aff = h5py.File(hdf5_aff_file, 'r') 28 | 29 | hdf5_raw_ds = pygt.normalize(np.asarray(hdf5_raw[hdf5_raw.keys()[0]]).astype(float32), -1, 1) 30 | hdf5_gt_ds = np.asarray(hdf5_gt[hdf5_gt.keys()[0]]).astype(float32) 31 | hdf5_aff_ds = np.asarray(hdf5_aff[hdf5_aff.keys()[0]]).astype(float32) 32 | 33 | datasets = [] 34 | for i in range(0,hdf5_raw_ds.shape[1]): 35 | dataset = {} 36 | dataset['data'] = hdf5_raw_ds[None, i, :] 37 | dataset['components'] = hdf5_gt_ds[None, i, :] 38 | dataset['label'] = hdf5_aff_ds[0:3, i, :] 39 | dataset['nhood'] = pygt.malis.mknhood2d() 40 | datasets += [dataset] 41 | 42 | #test_dataset = {} 43 | #test_dataset['data'] = hdf5_raw_ds 44 | #test_dataset['label'] = hdf5_aff_ds 45 | 46 | 47 | # Set train options 48 | class TrainOptions: 49 | loss_function = "euclid" 50 | loss_output_file = "log/loss.log" 51 | test_output_file = "log/test.log" 52 | test_interval = 4000 53 | scale_error = True 54 | training_method = "affinity" 55 | recompute_affinity = True 56 | train_device = 0 57 | test_device = 0 58 | test_net=None #'net_test.prototxt' 59 | 60 | 61 | options = TrainOptions() 62 | 63 | # Set solver options 64 | solver_config = pygt.caffe.SolverParameter() 65 | solver_config.train_net = 'net_train_euclid.prototxt' 66 | solver_config.base_lr = 0.0001 67 | solver_config.momentum = 0.99 68 | solver_config.weight_decay = 0.000005 69 | solver_config.lr_policy = 'inv' 70 | solver_config.gamma = 0.0001 71 | solver_config.power = 0.75 72 | solver_config.max_iter = 8000 73 | solver_config.snapshot = 2000 74 | solver_config.snapshot_prefix = 'net' 75 | solver_config.type = 'Adam' 76 | solver_config.display = 1 77 | 78 | # Set devices 79 | # pygt.caffe.enumerate_devices(False) 80 | pygt.caffe.set_devices((options.train_device,)) 81 | 82 | 83 | solverstates = pygt.getSolverStates(solver_config.snapshot_prefix); 84 | 85 | # First training method 86 | if (len(solverstates) == 0 or solverstates[-1][0] < solver_config.max_iter): 87 | solver, test_net = pygt.init_solver(solver_config, options) 88 | if (len(solverstates) > 0): 89 | solver.restore(solverstates[-1][1]) 90 | pygt.train(solver, test_net, datasets, [], options) 91 | 92 | 93 | 94 | 95 | 96 | 97 | 98 | --------------------------------------------------------------------------------