├── .gitignore
├── LICENSE
├── README.md
├── cfg
    ├── 0514
    │   ├── prune_0.7_keep_0.05_8x_yolov5l_v4.cfg
    │   ├── prune_0.8_keep_0.01_8x_yolov5l_v4.cfg
    │   ├── prune_0.8_keep_0.05_8x_yolov5l_v4.cfg
    │   ├── yolov5s_v4.cfg
    │   └── yolov5s_v4_hand.cfg
    ├── last_prune
    │   ├── prune_0.3_keep_0.05_8x_yolov5l_v4.cfg
    │   └── prune_0.8_keep_0.01_8x_yolov5l_v4.cfg
    ├── prune_0.7_keep_0.1_8x_yolov5l_v4.cfg
    ├── prune_0.8_keep_0.01_8x_yolov5l_v4.cfg
    └── yolov5l_v4.cfg
├── data
    ├── coco.data
    ├── coco.names
    ├── coco_128img.data
    ├── coco_128img.txt
    ├── get_coco_dataset.sh
    └── get_coco_dataset_gdrive.sh
├── models
    ├── __init__.py
    ├── common.py
    ├── experimental.py
    ├── export.py
    ├── hub
    │   ├── yolov3-spp.yaml
    │   ├── yolov5-fpn.yaml
    │   └── yolov5-panet.yaml
    ├── yolo.py
    └── yolov5s_v4.yaml
├── modelsori.py
├── prune_yolov5s.py
├── prune_yolov5s.sh
├── shortcut_prune_yolov5s.py
├── slim_prune_yolov5l_8x.py
├── slim_prune_yolov5s.py
├── slim_prune_yolov5s_8x.py
├── slim_prune_yolov5s_8x.sh
├── test.py
├── test_yolov5s.py
├── tk1_time.xls
└── utils
    ├── __init__.py
    ├── adabound.py
    ├── datasets.py
    ├── gcp.sh
    ├── general.py
    ├── google_utils.py
    ├── parse_config.py
    ├── prune_utils.py
    ├── torch_utils.py
    └── utils.py


/.gitignore:
--------------------------------------------------------------------------------
  1 | # Repo-specific GitIgnore ----------------------------------------------------------------------------------------------
  2 | *.jpg
  3 | *.jpeg
  4 | *.png
  5 | *.bmp
  6 | *.tif
  7 | *.tiff
  8 | *.heic
  9 | *.JPG
 10 | *.JPEG
 11 | *.PNG
 12 | *.BMP
 13 | *.TIF
 14 | *.TIFF
 15 | *.HEIC
 16 | *.mp4
 17 | *.mov
 18 | *.MOV
 19 | *.avi
 20 | *.data
 21 | *.json
 22 | 
 23 | #data
 24 | data/coco128
 25 | data/hand_dataset
 26 | 
 27 | 
 28 | # *.cfg
 29 | !cfg/yolov3*.cfg
 30 | 
 31 | storage.googleapis.com
 32 | runs/*
 33 | data/*
 34 | !data/samples/zidane.jpg
 35 | !data/samples/bus.jpg
 36 | !data/coco.names
 37 | !data/coco_paper.names
 38 | !data/coco.data
 39 | !data/coco_*.data
 40 | !data/coco_*.txt
 41 | !data/trainvalno5k.shapes
 42 | !data/*.sh
 43 | 
 44 | pycocotools/*
 45 | results*.txt
 46 | gcp_test*.sh
 47 | 
 48 | # MATLAB GitIgnore -----------------------------------------------------------------------------------------------------
 49 | *.m~
 50 | *.mat
 51 | !targets*.mat
 52 | 
 53 | # Neural Network weights -----------------------------------------------------------------------------------------------
 54 | *.weights
 55 | *.pt
 56 | *.onnx
 57 | *.mlmodel
 58 | *.torchscript
 59 | darknet53.conv.74
 60 | yolov3-tiny.conv.15
 61 | 
 62 | # GitHub Python GitIgnore ----------------------------------------------------------------------------------------------
 63 | # Byte-compiled / optimized / DLL files
 64 | __pycache__/
 65 | *.py[cod]
 66 | *$py.class
 67 | 
 68 | # C extensions
 69 | *.so
 70 | 
 71 | # Distribution / packaging
 72 | .Python
 73 | env/
 74 | build/
 75 | develop-eggs/
 76 | dist/
 77 | downloads/
 78 | eggs/
 79 | .eggs/
 80 | lib/
 81 | lib64/
 82 | parts/
 83 | sdist/
 84 | var/
 85 | wheels/
 86 | *.egg-info/
 87 | .installed.cfg
 88 | *.egg
 89 | 
 90 | # PyInstaller
 91 | #  Usually these files are written by a python script from a template
 92 | #  before PyInstaller builds the exe, so as to inject date/other infos into it.
 93 | *.manifest
 94 | *.spec
 95 | 
 96 | # Installer logs
 97 | pip-log.txt
 98 | pip-delete-this-directory.txt
 99 | 
100 | # Unit test / coverage reports
101 | htmlcov/
102 | .tox/
103 | .coverage
104 | .coverage.*
105 | .cache
106 | nosetests.xml
107 | coverage.xml
108 | *.cover
109 | .hypothesis/
110 | 
111 | # Translations
112 | *.mo
113 | *.pot
114 | 
115 | # Django stuff:
116 | *.log
117 | local_settings.py
118 | 
119 | # Flask stuff:
120 | instance/
121 | .webassets-cache
122 | 
123 | # Scrapy stuff:
124 | .scrapy
125 | 
126 | # Sphinx documentation
127 | docs/_build/
128 | 
129 | # PyBuilder
130 | target/
131 | 
132 | # Jupyter Notebook
133 | .ipynb_checkpoints
134 | 
135 | # pyenv
136 | .python-version
137 | 
138 | # celery beat schedule file
139 | celerybeat-schedule
140 | 
141 | # SageMath parsed files
142 | *.sage.py
143 | 
144 | # dotenv
145 | .env
146 | 
147 | # virtualenv
148 | .venv
149 | venv/
150 | ENV/
151 | 
152 | # Spyder project settings
153 | .spyderproject
154 | .spyproject
155 | 
156 | # Rope project settings
157 | .ropeproject
158 | 
159 | # mkdocs documentation
160 | /site
161 | 
162 | # mypy
163 | .mypy_cache/
164 | 
165 | 
166 | # https://github.com/github/gitignore/blob/master/Global/macOS.gitignore -----------------------------------------------
167 | 
168 | # General
169 | .DS_Store
170 | .AppleDouble
171 | .LSOverride
172 | 
173 | # Icon must end with two \r
174 | Icon
175 | Icon?
176 | 
177 | # Thumbnails
178 | ._*
179 | 
180 | # Files that might appear in the root of a volume
181 | .DocumentRevisions-V100
182 | .fseventsd
183 | .Spotlight-V100
184 | .TemporaryItems
185 | .Trashes
186 | .VolumeIcon.icns
187 | .com.apple.timemachine.donotpresent
188 | 
189 | # Directories potentially created on remote AFP share
190 | .AppleDB
191 | .AppleDesktop
192 | Network Trash Folder
193 | Temporary Items
194 | .apdisk
195 | 
196 | 
197 | # https://github.com/github/gitignore/blob/master/Global/JetBrains.gitignore
198 | # Covers JetBrains IDEs: IntelliJ, RubyMine, PhpStorm, AppCode, PyCharm, CLion, Android Studio and WebStorm
199 | # Reference: https://intellij-support.jetbrains.com/hc/en-us/articles/206544839
200 | 
201 | # User-specific stuff:
202 | .idea/*
203 | .idea/**/workspace.xml
204 | .idea/**/tasks.xml
205 | .idea/dictionaries
206 | .html  # Bokeh Plots
207 | .pg  # TensorFlow Frozen Graphs
208 | .avi # videos
209 | 
210 | # Sensitive or high-churn files:
211 | .idea/**/dataSources/
212 | .idea/**/dataSources.ids
213 | .idea/**/dataSources.local.xml
214 | .idea/**/sqlDataSources.xml
215 | .idea/**/dynamic.xml
216 | .idea/**/uiDesigner.xml
217 | 
218 | # Gradle:
219 | .idea/**/gradle.xml
220 | .idea/**/libraries
221 | 
222 | # CMake
223 | cmake-build-debug/
224 | cmake-build-release/
225 | 
226 | # Mongo Explorer plugin:
227 | .idea/**/mongoSettings.xml
228 | 
229 | ## File-based project format:
230 | *.iws
231 | 
232 | ## Plugin-specific files:
233 | 
234 | # IntelliJ
235 | out/
236 | 
237 | # mpeltonen/sbt-idea plugin
238 | .idea_modules/
239 | 
240 | # JIRA plugin
241 | atlassian-ide-plugin.xml
242 | 
243 | # Cursive Clojure plugin
244 | .idea/replstate.xml
245 | 
246 | # Crashlytics plugin (for Android Studio and IntelliJ)
247 | com_crashlytics_export_strings.xml
248 | crashlytics.properties
249 | crashlytics-build.properties
250 | fabric.properties
251 | 
252 | tensorboard/
253 | 


--------------------------------------------------------------------------------
/LICENSE:
--------------------------------------------------------------------------------
  1 |                                  Apache License
  2 |                            Version 2.0, January 2004
  3 |                         http://www.apache.org/licenses/
  4 | 
  5 |    TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
  6 | 
  7 |    1. Definitions.
  8 | 
  9 |       "License" shall mean the terms and conditions for use, reproduction,
 10 |       and distribution as defined by Sections 1 through 9 of this document.
 11 | 
 12 |       "Licensor" shall mean the copyright owner or entity authorized by
 13 |       the copyright owner that is granting the License.
 14 | 
 15 |       "Legal Entity" shall mean the union of the acting entity and all
 16 |       other entities that control, are controlled by, or are under common
 17 |       control with that entity. For the purposes of this definition,
 18 |       "control" means (i) the power, direct or indirect, to cause the
 19 |       direction or management of such entity, whether by contract or
 20 |       otherwise, or (ii) ownership of fifty percent (50%) or more of the
 21 |       outstanding shares, or (iii) beneficial ownership of such entity.
 22 | 
 23 |       "You" (or "Your") shall mean an individual or Legal Entity
 24 |       exercising permissions granted by this License.
 25 | 
 26 |       "Source" form shall mean the preferred form for making modifications,
 27 |       including but not limited to software source code, documentation
 28 |       source, and configuration files.
 29 | 
 30 |       "Object" form shall mean any form resulting from mechanical
 31 |       transformation or translation of a Source form, including but
 32 |       not limited to compiled object code, generated documentation,
 33 |       and conversions to other media types.
 34 | 
 35 |       "Work" shall mean the work of authorship, whether in Source or
 36 |       Object form, made available under the License, as indicated by a
 37 |       copyright notice that is included in or attached to the work
 38 |       (an example is provided in the Appendix below).
 39 | 
 40 |       "Derivative Works" shall mean any work, whether in Source or Object
 41 |       form, that is based on (or derived from) the Work and for which the
 42 |       editorial revisions, annotations, elaborations, or other modifications
 43 |       represent, as a whole, an original work of authorship. For the purposes
 44 |       of this License, Derivative Works shall not include works that remain
 45 |       separable from, or merely link (or bind by name) to the interfaces of,
 46 |       the Work and Derivative Works thereof.
 47 | 
 48 |       "Contribution" shall mean any work of authorship, including
 49 |       the original version of the Work and any modifications or additions
 50 |       to that Work or Derivative Works thereof, that is intentionally
 51 |       submitted to Licensor for inclusion in the Work by the copyright owner
 52 |       or by an individual or Legal Entity authorized to submit on behalf of
 53 |       the copyright owner. For the purposes of this definition, "submitted"
 54 |       means any form of electronic, verbal, or written communication sent
 55 |       to the Licensor or its representatives, including but not limited to
 56 |       communication on electronic mailing lists, source code control systems,
 57 |       and issue tracking systems that are managed by, or on behalf of, the
 58 |       Licensor for the purpose of discussing and improving the Work, but
 59 |       excluding communication that is conspicuously marked or otherwise
 60 |       designated in writing by the copyright owner as "Not a Contribution."
 61 | 
 62 |       "Contributor" shall mean Licensor and any individual or Legal Entity
 63 |       on behalf of whom a Contribution has been received by Licensor and
 64 |       subsequently incorporated within the Work.
 65 | 
 66 |    2. Grant of Copyright License. Subject to the terms and conditions of
 67 |       this License, each Contributor hereby grants to You a perpetual,
 68 |       worldwide, non-exclusive, no-charge, royalty-free, irrevocable
 69 |       copyright license to reproduce, prepare Derivative Works of,
 70 |       publicly display, publicly perform, sublicense, and distribute the
 71 |       Work and such Derivative Works in Source or Object form.
 72 | 
 73 |    3. Grant of Patent License. Subject to the terms and conditions of
 74 |       this License, each Contributor hereby grants to You a perpetual,
 75 |       worldwide, non-exclusive, no-charge, royalty-free, irrevocable
 76 |       (except as stated in this section) patent license to make, have made,
 77 |       use, offer to sell, sell, import, and otherwise transfer the Work,
 78 |       where such license applies only to those patent claims licensable
 79 |       by such Contributor that are necessarily infringed by their
 80 |       Contribution(s) alone or by combination of their Contribution(s)
 81 |       with the Work to which such Contribution(s) was submitted. If You
 82 |       institute patent litigation against any entity (including a
 83 |       cross-claim or counterclaim in a lawsuit) alleging that the Work
 84 |       or a Contribution incorporated within the Work constitutes direct
 85 |       or contributory patent infringement, then any patent licenses
 86 |       granted to You under this License for that Work shall terminate
 87 |       as of the date such litigation is filed.
 88 | 
 89 |    4. Redistribution. You may reproduce and distribute copies of the
 90 |       Work or Derivative Works thereof in any medium, with or without
 91 |       modifications, and in Source or Object form, provided that You
 92 |       meet the following conditions:
 93 | 
 94 |       (a) You must give any other recipients of the Work or
 95 |           Derivative Works a copy of this License; and
 96 | 
 97 |       (b) You must cause any modified files to carry prominent notices
 98 |           stating that You changed the files; and
 99 | 
100 |       (c) You must retain, in the Source form of any Derivative Works
101 |           that You distribute, all copyright, patent, trademark, and
102 |           attribution notices from the Source form of the Work,
103 |           excluding those notices that do not pertain to any part of
104 |           the Derivative Works; and
105 | 
106 |       (d) If the Work includes a "NOTICE" text file as part of its
107 |           distribution, then any Derivative Works that You distribute must
108 |           include a readable copy of the attribution notices contained
109 |           within such NOTICE file, excluding those notices that do not
110 |           pertain to any part of the Derivative Works, in at least one
111 |           of the following places: within a NOTICE text file distributed
112 |           as part of the Derivative Works; within the Source form or
113 |           documentation, if provided along with the Derivative Works; or,
114 |           within a display generated by the Derivative Works, if and
115 |           wherever such third-party notices normally appear. The contents
116 |           of the NOTICE file are for informational purposes only and
117 |           do not modify the License. You may add Your own attribution
118 |           notices within Derivative Works that You distribute, alongside
119 |           or as an addendum to the NOTICE text from the Work, provided
120 |           that such additional attribution notices cannot be construed
121 |           as modifying the License.
122 | 
123 |       You may add Your own copyright statement to Your modifications and
124 |       may provide additional or different license terms and conditions
125 |       for use, reproduction, or distribution of Your modifications, or
126 |       for any such Derivative Works as a whole, provided Your use,
127 |       reproduction, and distribution of the Work otherwise complies with
128 |       the conditions stated in this License.
129 | 
130 |    5. Submission of Contributions. Unless You explicitly state otherwise,
131 |       any Contribution intentionally submitted for inclusion in the Work
132 |       by You to the Licensor shall be under the terms and conditions of
133 |       this License, without any additional terms or conditions.
134 |       Notwithstanding the above, nothing herein shall supersede or modify
135 |       the terms of any separate license agreement you may have executed
136 |       with Licensor regarding such Contributions.
137 | 
138 |    6. Trademarks. This License does not grant permission to use the trade
139 |       names, trademarks, service marks, or product names of the Licensor,
140 |       except as required for reasonable and customary use in describing the
141 |       origin of the Work and reproducing the content of the NOTICE file.
142 | 
143 |    7. Disclaimer of Warranty. Unless required by applicable law or
144 |       agreed to in writing, Licensor provides the Work (and each
145 |       Contributor provides its Contributions) on an "AS IS" BASIS,
146 |       WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
147 |       implied, including, without limitation, any warranties or conditions
148 |       of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
149 |       PARTICULAR PURPOSE. You are solely responsible for determining the
150 |       appropriateness of using or redistributing the Work and assume any
151 |       risks associated with Your exercise of permissions under this License.
152 | 
153 |    8. Limitation of Liability. In no event and under no legal theory,
154 |       whether in tort (including negligence), contract, or otherwise,
155 |       unless required by applicable law (such as deliberate and grossly
156 |       negligent acts) or agreed to in writing, shall any Contributor be
157 |       liable to You for damages, including any direct, indirect, special,
158 |       incidental, or consequential damages of any character arising as a
159 |       result of this License or out of the use or inability to use the
160 |       Work (including but not limited to damages for loss of goodwill,
161 |       work stoppage, computer failure or malfunction, or any and all
162 |       other commercial damages or losses), even if such Contributor
163 |       has been advised of the possibility of such damages.
164 | 
165 |    9. Accepting Warranty or Additional Liability. While redistributing
166 |       the Work or Derivative Works thereof, You may choose to offer,
167 |       and charge a fee for, acceptance of support, warranty, indemnity,
168 |       or other liability obligations and/or rights consistent with this
169 |       License. However, in accepting such obligations, You may act only
170 |       on Your own behalf and on Your sole responsibility, not on behalf
171 |       of any other Contributor, and only if You agree to indemnify,
172 |       defend, and hold each Contributor harmless for any liability
173 |       incurred by, or claims asserted against, such Contributor by reason
174 |       of your accepting any such warranty or additional liability.
175 | 
176 |    END OF TERMS AND CONDITIONS
177 | 
178 |    APPENDIX: How to apply the Apache License to your work.
179 | 
180 |       To apply the Apache License to your work, attach the following
181 |       boilerplate notice, with the fields enclosed by brackets "[]"
182 |       replaced with your own identifying information. (Don't include
183 |       the brackets!)  The text should be enclosed in the appropriate
184 |       comment syntax for the file format. We also recommend that a
185 |       file or class name and description of purpose be included on the
186 |       same "printed page" as the copyright notice for easier
187 |       identification within third-party archives.
188 | 
189 |    Copyright [yyyy] [name of copyright owner]
190 | 
191 |    Licensed under the Apache License, Version 2.0 (the "License");
192 |    you may not use this file except in compliance with the License.
193 |    You may obtain a copy of the License at
194 | 
195 |        http://www.apache.org/licenses/LICENSE-2.0
196 | 
197 |    Unless required by applicable law or agreed to in writing, software
198 |    distributed under the License is distributed on an "AS IS" BASIS,
199 |    WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
200 |    See the License for the specific language governing permissions and
201 |    limitations under the License.
202 | 


--------------------------------------------------------------------------------
/README.md:
--------------------------------------------------------------------------------
 1 | # yolov5_prune
 2 | 本项目作为 [ZJU-lishuang](https://github.com/ZJU-lishuang)/**[yolov5_prune](https://github.com/ZJU-lishuang/yolov5_prune)**的补充，实现了l模型的剪枝。其余过程都可参考lishuang的repo。
 3 | 
 4 | 本项目基于[tanluren/yolov3-channel-and-layer-pruning](https://github.com/tanluren/yolov3-channel-and-layer-pruning)实现，将项目扩展到yolov5上。
 5 | 
 6 | 项目的基本流程是，使用[ultralytics/yolov5](https://github.com/ultralytics/yolov5)训练自己的数据集，在模型性能达到要求但速度未达到要求时，对模型进行剪枝。首先是稀疏化训练，稀疏化训练很重要，如果模型稀疏度不够，剪枝比例过大会导致剪枝后的模型map接近0。剪枝完成后对模型进行微调回复精度。
 7 | 
 8 | 本项目使用的yolov5为第四版本。
 9 | yolov5第三版本参考[yolov5-v3-prune](https://github.com/ZJU-lishuang/yolov5_prune/tree/v3)
10 | yolov5第二版本参考[yolov5-v2-prune](https://github.com/ZJU-lishuang/yolov5_prune/tree/v2)
11 | 
12 | TODO： 增加m,l,x的模型剪枝，如果有时间的话。>-<
13 | 
14 | PS：在开源数据集和不能开源的数据集上模型均剪枝成功。
15 | 
16 | ## 实例流程
17 | 数据集下载[dataset](http://www.robots.ox.ac.uk/~vgg/data/hands/downloads/hand_dataset.tar.gz)<br>
18 | ### STEP1:基础训练 
19 | 附件：[训练记录](https://drive.google.com/drive/folders/1v0HZYBhU6d4M2hvEfjia76wYbQlaFz_f?usp=sharing)<br>
20 | ### STEP2:稀疏训练     
21 | 附件：[稀疏训练记录](https://drive.google.com/drive/folders/1tJaeSOzQlyrx1l22hhop8G3ZuKshm8rp?usp=sharing)<br>
22 | ### STEP3:八倍通道剪枝  
23 | 附件：[剪枝后模型](https://drive.google.com/drive/folders/1V5nA6oGXX43bagpO3cJIFpI0zjAOzt0p?usp=sharing)<br>
24 | ### STEP4:微调finetune 
25 | 附件：[微调训练记录](https://drive.google.com/drive/folders/1vT_pN_XlMBniF9YXaPj2KeCNZitxYFLA?usp=sharing)<br>
26 | ### STEP4:微调finetune，使用蒸馏技术优化模型，效果由于单纯的微调模型 
27 | 附件：[微调蒸馏训练记录](https://drive.google.com/drive/folders/1T3SGh0FjyjxDckFcKVxpxQHF2XzZ-gfN?usp=sharing)<br>
28 | 
29 | ## 剪枝步骤
30 | #### STEP1:基础训练
31 | [yolov5](https://github.com/ZJU-lishuang/yolov5-v4) <br>
32 | 示例代码 <br>
33 | ```
34 | python train.py --img 640 --batch 8 --epochs 50 --weights weights/yolov5s_v4.pt --data data/coco_hand.yaml --cfg models/yolov5s.yaml --name s_hand
35 | ```
36 | 
37 | #### STEP2:稀疏训练
38 | --prune 0 适用于通道剪枝策略一，--prune 1 适用于其他剪枝策略。<br>
39 | [yolov5](https://github.com/ZJU-lishuang/yolov5-v4)<br>
40 | 示例代码<br>
41 | ```
42 | python train_sparsity.py --img 640 --batch 8 --epochs 50 --data data/coco_hand.yaml --cfg models/yolov5s.yaml --weights runs/train/s_hand/weights/last.pt --name s_hand_sparsity -sr --s 0.001 --prune 1
43 | ```
44 | 
45 | #### STEP3:通道剪枝策略一
46 | 不对shortcut直连的层进行剪枝，避免维度处理。<br>
47 | ```
48 | python prune_yolov5s.py --cfg cfg/yolov5s.cfg --data data/fangweisui.data --weights weights/yolov5s_prune0.pt --percent 0.8
49 | ```
50 | 
51 | #### STEP3:通道剪枝策略二
52 | 对shortcut层也进行了剪枝，剪枝采用每组shortcut中第一个卷积层的mask。<br>
53 | ```
54 | python shortcut_prune_yolov5s.py --cfg cfg/yolov5s.cfg --data data/fangweisui.data --weights weights/yolov5s_prune1.pt --percent 0.3
55 | ```
56 | 
57 | #### STEP3:通道剪枝策略三
58 | 先以全局阈值找出各卷积层的mask，然后对于每组shortcut，它将相连的各卷积层的剪枝mask取并集，用merge后的mask进行剪枝。<br>
59 | ```
60 | python slim_prune_yolov5s.py --cfg cfg/yolov5s.cfg --data data/fangweisui.data --weights weights/yolov5s_prune1.pt --global_percent 0.8 --layer_keep 0.01
61 | ```
62 | 
63 | #### STEP3:八倍通道剪枝
64 | 在硬件部署上发现，模型剪枝率相同时，通道数为8的倍数速度最快。（采坑：需要将硬件性能开启到最大）<br>
65 | 示例代码<br>
66 | ```
67 | python slim_prune_yolov5s_8x.py --cfg cfg/yolov5s_v4_hand.cfg --data data/oxfordhand.data --weights weights/last_v4s.pt --global_percent 0.5 --layer_keep 0.01 --img_size 640
68 | ```
69 | 
70 | #### STEP4:微调finetune
71 | [yolov5](https://github.com/ZJU-lishuang/yolov5-v4)<br>
72 | 示例代码<br>
73 | ```
74 | python prune_finetune.py --img 640 --batch 8 --epochs 50 --data data/coco_hand.yaml --cfg ./cfg/prune_0.5_keep_0.01_8x_yolov5s_v4_hand.cfg --weights ./weights/prune_0.5_keep_0.01_8x_last_v4s.pt --name s_hand_finetune
75 | ```
76 | 
77 | #### STEP4:微调finetune，使用蒸馏技术优化模型
78 | [yolov5](https://github.com/ZJU-lishuang/yolov5-v4)<br>
79 | 示例代码<br>
80 | ```
81 | python prune_finetune.py --img 640 --batch 8 --epochs 50 --data data/coco_hand.yaml --cfg ./cfg/prune_0.5_keep_0.01_8x_yolov5s_v4_hand.cfg --weights ./weights/prune_0.5_keep_0.01_8x_last_v4s.pt --name s_hand_finetune_distill --distill
82 | ```
83 | 
84 | #### STEP5:剪枝后模型推理
85 | [yolov5](https://github.com/ZJU-lishuang/yolov5-v4)<br>
86 | 示例代码<br>
87 | ```shell
88 | python prune_detect.py --weights weights/last_s_hand_finetune.pt --img  640 --conf 0.7 --save-txt --source inference/images
89 | ```
90 | 
91 | 
92 | 


--------------------------------------------------------------------------------
/cfg/0514/prune_0.7_keep_0.05_8x_yolov5l_v4.cfg:
--------------------------------------------------------------------------------
   1 | [net]
   2 | batch=64
   3 | subdivisions=8
   4 | width=416
   5 | height=416
   6 | channels=3
   7 | momentum=0.949
   8 | decay=0.0005
   9 | angle=0
  10 | saturation=1.5
  11 | exposure=1.5
  12 | hue=.1
  13 | learning_rate=0.00261
  14 | burn_in=1000
  15 | max_batches=500500
  16 | policy=steps
  17 | steps=400000,450000
  18 | scales=.1,.1
  19 | mosaic=1
  20 | 
  21 | [focus]
  22 | filters=12
  23 | 
  24 | [convolutional]
  25 | batch_normalize=1
  26 | filters=56
  27 | size=3
  28 | stride=1
  29 | pad=1
  30 | activation=leaky
  31 | 
  32 | [convolutional]
  33 | batch_normalize=1
  34 | filters=128
  35 | size=3
  36 | stride=2
  37 | pad=1
  38 | activation=leaky
  39 | 
  40 | [convolutional]
  41 | batch_normalize=1
  42 | filters=64
  43 | size=1
  44 | stride=1
  45 | pad=1
  46 | activation=leaky
  47 | 
  48 | [route]
  49 | layers=-2
  50 | 
  51 | [convolutional]
  52 | batch_normalize=1
  53 | filters=64
  54 | size=1
  55 | stride=1
  56 | pad=1
  57 | activation=leaky
  58 | 
  59 | [convolutional]
  60 | batch_normalize=1
  61 | filters=64
  62 | size=1
  63 | stride=1
  64 | pad=1
  65 | activation=leaky
  66 | 
  67 | [convolutional]
  68 | batch_normalize=1
  69 | filters=64
  70 | size=3
  71 | stride=1
  72 | pad=1
  73 | activation=leaky
  74 | 
  75 | [shortcut]
  76 | from=-3
  77 | activation=linear
  78 | 
  79 | [convolutional]
  80 | batch_normalize=1
  81 | filters=64
  82 | size=1
  83 | stride=1
  84 | pad=1
  85 | activation=leaky
  86 | 
  87 | [convolutional]
  88 | batch_normalize=1
  89 | filters=64
  90 | size=3
  91 | stride=1
  92 | pad=1
  93 | activation=leaky
  94 | 
  95 | [shortcut]
  96 | from=-3
  97 | activation=linear
  98 | 
  99 | [convolutional]
 100 | batch_normalize=1
 101 | filters=64
 102 | size=1
 103 | stride=1
 104 | pad=1
 105 | activation=leaky
 106 | 
 107 | [convolutional]
 108 | batch_normalize=1
 109 | filters=64
 110 | size=3
 111 | stride=1
 112 | pad=1
 113 | activation=leaky
 114 | 
 115 | [shortcut]
 116 | from=-3
 117 | activation=linear
 118 | 
 119 | [route]
 120 | layers=-1,-12
 121 | 
 122 | [convolutional]
 123 | batch_normalize=1
 124 | filters=128
 125 | size=1
 126 | stride=1
 127 | pad=1
 128 | activation=leaky
 129 | 
 130 | [convolutional]
 131 | batch_normalize=1
 132 | filters=256
 133 | size=3
 134 | stride=2
 135 | pad=1
 136 | activation=leaky
 137 | 
 138 | [convolutional]
 139 | batch_normalize=1
 140 | filters=128
 141 | size=1
 142 | stride=1
 143 | pad=1
 144 | activation=leaky
 145 | 
 146 | [route]
 147 | layers=-2
 148 | 
 149 | [convolutional]
 150 | batch_normalize=1
 151 | filters=128
 152 | size=1
 153 | stride=1
 154 | pad=1
 155 | activation=leaky
 156 | 
 157 | [convolutional]
 158 | batch_normalize=1
 159 | filters=72
 160 | size=1
 161 | stride=1
 162 | pad=1
 163 | activation=leaky
 164 | 
 165 | [convolutional]
 166 | batch_normalize=1
 167 | filters=128
 168 | size=3
 169 | stride=1
 170 | pad=1
 171 | activation=leaky
 172 | 
 173 | [shortcut]
 174 | from=-3
 175 | activation=linear
 176 | 
 177 | [convolutional]
 178 | batch_normalize=1
 179 | filters=64
 180 | size=1
 181 | stride=1
 182 | pad=1
 183 | activation=leaky
 184 | 
 185 | [convolutional]
 186 | batch_normalize=1
 187 | filters=128
 188 | size=3
 189 | stride=1
 190 | pad=1
 191 | activation=leaky
 192 | 
 193 | [shortcut]
 194 | from=-3
 195 | activation=linear
 196 | 
 197 | [convolutional]
 198 | batch_normalize=1
 199 | filters=48
 200 | size=1
 201 | stride=1
 202 | pad=1
 203 | activation=leaky
 204 | 
 205 | [convolutional]
 206 | batch_normalize=1
 207 | filters=128
 208 | size=3
 209 | stride=1
 210 | pad=1
 211 | activation=leaky
 212 | 
 213 | [shortcut]
 214 | from=-3
 215 | activation=linear
 216 | 
 217 | [convolutional]
 218 | batch_normalize=1
 219 | filters=40
 220 | size=1
 221 | stride=1
 222 | pad=1
 223 | activation=leaky
 224 | 
 225 | [convolutional]
 226 | batch_normalize=1
 227 | filters=128
 228 | size=3
 229 | stride=1
 230 | pad=1
 231 | activation=leaky
 232 | 
 233 | [shortcut]
 234 | from=-3
 235 | activation=linear
 236 | 
 237 | [convolutional]
 238 | batch_normalize=1
 239 | filters=64
 240 | size=1
 241 | stride=1
 242 | pad=1
 243 | activation=leaky
 244 | 
 245 | [convolutional]
 246 | batch_normalize=1
 247 | filters=128
 248 | size=3
 249 | stride=1
 250 | pad=1
 251 | activation=leaky
 252 | 
 253 | [shortcut]
 254 | from=-3
 255 | activation=linear
 256 | 
 257 | [convolutional]
 258 | batch_normalize=1
 259 | filters=48
 260 | size=1
 261 | stride=1
 262 | pad=1
 263 | activation=leaky
 264 | 
 265 | [convolutional]
 266 | batch_normalize=1
 267 | filters=128
 268 | size=3
 269 | stride=1
 270 | pad=1
 271 | activation=leaky
 272 | 
 273 | [shortcut]
 274 | from=-3
 275 | activation=linear
 276 | 
 277 | [convolutional]
 278 | batch_normalize=1
 279 | filters=80
 280 | size=1
 281 | stride=1
 282 | pad=1
 283 | activation=leaky
 284 | 
 285 | [convolutional]
 286 | batch_normalize=1
 287 | filters=128
 288 | size=3
 289 | stride=1
 290 | pad=1
 291 | activation=leaky
 292 | 
 293 | [shortcut]
 294 | from=-3
 295 | activation=linear
 296 | 
 297 | [convolutional]
 298 | batch_normalize=1
 299 | filters=88
 300 | size=1
 301 | stride=1
 302 | pad=1
 303 | activation=leaky
 304 | 
 305 | [convolutional]
 306 | batch_normalize=1
 307 | filters=128
 308 | size=3
 309 | stride=1
 310 | pad=1
 311 | activation=leaky
 312 | 
 313 | [shortcut]
 314 | from=-3
 315 | activation=linear
 316 | 
 317 | [convolutional]
 318 | batch_normalize=1
 319 | filters=72
 320 | size=1
 321 | stride=1
 322 | pad=1
 323 | activation=leaky
 324 | 
 325 | [convolutional]
 326 | batch_normalize=1
 327 | filters=128
 328 | size=3
 329 | stride=1
 330 | pad=1
 331 | activation=leaky
 332 | 
 333 | [shortcut]
 334 | from=-3
 335 | activation=linear
 336 | 
 337 | [route]
 338 | layers=-1,-30
 339 | 
 340 | [convolutional]
 341 | batch_normalize=1
 342 | filters=256
 343 | size=1
 344 | stride=1
 345 | pad=1
 346 | activation=leaky
 347 | 
 348 | [convolutional]
 349 | batch_normalize=1
 350 | filters=328
 351 | size=3
 352 | stride=2
 353 | pad=1
 354 | activation=leaky
 355 | 
 356 | [convolutional]
 357 | batch_normalize=1
 358 | filters=232
 359 | size=1
 360 | stride=1
 361 | pad=1
 362 | activation=leaky
 363 | 
 364 | [route]
 365 | layers=-2
 366 | 
 367 | [convolutional]
 368 | batch_normalize=1
 369 | filters=254
 370 | size=1
 371 | stride=1
 372 | pad=1
 373 | activation=leaky
 374 | 
 375 | [convolutional]
 376 | batch_normalize=1
 377 | filters=80
 378 | size=1
 379 | stride=1
 380 | pad=1
 381 | activation=leaky
 382 | 
 383 | [convolutional]
 384 | batch_normalize=1
 385 | filters=254
 386 | size=3
 387 | stride=1
 388 | pad=1
 389 | activation=leaky
 390 | 
 391 | [shortcut]
 392 | from=-3
 393 | activation=linear
 394 | 
 395 | [convolutional]
 396 | batch_normalize=1
 397 | filters=16
 398 | size=1
 399 | stride=1
 400 | pad=1
 401 | activation=leaky
 402 | 
 403 | [convolutional]
 404 | batch_normalize=1
 405 | filters=254
 406 | size=3
 407 | stride=1
 408 | pad=1
 409 | activation=leaky
 410 | 
 411 | [shortcut]
 412 | from=-3
 413 | activation=linear
 414 | 
 415 | [convolutional]
 416 | batch_normalize=1
 417 | filters=24
 418 | size=1
 419 | stride=1
 420 | pad=1
 421 | activation=leaky
 422 | 
 423 | [convolutional]
 424 | batch_normalize=1
 425 | filters=254
 426 | size=3
 427 | stride=1
 428 | pad=1
 429 | activation=leaky
 430 | 
 431 | [shortcut]
 432 | from=-3
 433 | activation=linear
 434 | 
 435 | [convolutional]
 436 | batch_normalize=1
 437 | filters=32
 438 | size=1
 439 | stride=1
 440 | pad=1
 441 | activation=leaky
 442 | 
 443 | [convolutional]
 444 | batch_normalize=1
 445 | filters=254
 446 | size=3
 447 | stride=1
 448 | pad=1
 449 | activation=leaky
 450 | 
 451 | [shortcut]
 452 | from=-3
 453 | activation=linear
 454 | 
 455 | [convolutional]
 456 | batch_normalize=1
 457 | filters=32
 458 | size=1
 459 | stride=1
 460 | pad=1
 461 | activation=leaky
 462 | 
 463 | [convolutional]
 464 | batch_normalize=1
 465 | filters=254
 466 | size=3
 467 | stride=1
 468 | pad=1
 469 | activation=leaky
 470 | 
 471 | [shortcut]
 472 | from=-3
 473 | activation=linear
 474 | 
 475 | [convolutional]
 476 | batch_normalize=1
 477 | filters=40
 478 | size=1
 479 | stride=1
 480 | pad=1
 481 | activation=leaky
 482 | 
 483 | [convolutional]
 484 | batch_normalize=1
 485 | filters=254
 486 | size=3
 487 | stride=1
 488 | pad=1
 489 | activation=leaky
 490 | 
 491 | [shortcut]
 492 | from=-3
 493 | activation=linear
 494 | 
 495 | [convolutional]
 496 | batch_normalize=1
 497 | filters=24
 498 | size=1
 499 | stride=1
 500 | pad=1
 501 | activation=leaky
 502 | 
 503 | [convolutional]
 504 | batch_normalize=1
 505 | filters=254
 506 | size=3
 507 | stride=1
 508 | pad=1
 509 | activation=leaky
 510 | 
 511 | [shortcut]
 512 | from=-3
 513 | activation=linear
 514 | 
 515 | [convolutional]
 516 | batch_normalize=1
 517 | filters=32
 518 | size=1
 519 | stride=1
 520 | pad=1
 521 | activation=leaky
 522 | 
 523 | [convolutional]
 524 | batch_normalize=1
 525 | filters=254
 526 | size=3
 527 | stride=1
 528 | pad=1
 529 | activation=leaky
 530 | 
 531 | [shortcut]
 532 | from=-3
 533 | activation=linear
 534 | 
 535 | [convolutional]
 536 | batch_normalize=1
 537 | filters=32
 538 | size=1
 539 | stride=1
 540 | pad=1
 541 | activation=leaky
 542 | 
 543 | [convolutional]
 544 | batch_normalize=1
 545 | filters=254
 546 | size=3
 547 | stride=1
 548 | pad=1
 549 | activation=leaky
 550 | 
 551 | [shortcut]
 552 | from=-3
 553 | activation=linear
 554 | 
 555 | [route]
 556 | layers=-1,-30
 557 | 
 558 | [convolutional]
 559 | batch_normalize=1
 560 | filters=224
 561 | size=1
 562 | stride=1
 563 | pad=1
 564 | activation=leaky
 565 | 
 566 | [convolutional]
 567 | batch_normalize=1
 568 | filters=128
 569 | size=3
 570 | stride=2
 571 | pad=1
 572 | activation=leaky
 573 | 
 574 | [convolutional]
 575 | batch_normalize=1
 576 | filters=512
 577 | size=1
 578 | stride=1
 579 | pad=1
 580 | activation=leaky
 581 | 
 582 | [maxpool]
 583 | stride=1
 584 | size=5
 585 | 
 586 | [route]
 587 | layers=-2
 588 | 
 589 | [maxpool]
 590 | stride=1
 591 | size=9
 592 | 
 593 | [route]
 594 | layers=-4
 595 | 
 596 | [maxpool]
 597 | stride=1
 598 | size=13
 599 | 
 600 | [route]
 601 | layers=-6,-5,-3,-1
 602 | 
 603 | [convolutional]
 604 | batch_normalize=1
 605 | filters=112
 606 | size=1
 607 | stride=1
 608 | pad=1
 609 | activation=leaky
 610 | 
 611 | [convolutional]
 612 | batch_normalize=1
 613 | filters=56
 614 | size=1
 615 | stride=1
 616 | pad=1
 617 | activation=leaky
 618 | 
 619 | [route]
 620 | layers=-2
 621 | 
 622 | [convolutional]
 623 | batch_normalize=1
 624 | filters=40
 625 | size=1
 626 | stride=1
 627 | pad=1
 628 | activation=leaky
 629 | 
 630 | [convolutional]
 631 | batch_normalize=1
 632 | filters=40
 633 | size=1
 634 | stride=1
 635 | pad=1
 636 | activation=leaky
 637 | 
 638 | [convolutional]
 639 | batch_normalize=1
 640 | filters=40
 641 | size=3
 642 | stride=1
 643 | pad=1
 644 | activation=leaky
 645 | 
 646 | [convolutional]
 647 | batch_normalize=1
 648 | filters=32
 649 | size=1
 650 | stride=1
 651 | pad=1
 652 | activation=leaky
 653 | 
 654 | [convolutional]
 655 | batch_normalize=1
 656 | filters=48
 657 | size=3
 658 | stride=1
 659 | pad=1
 660 | activation=leaky
 661 | 
 662 | [convolutional]
 663 | batch_normalize=1
 664 | filters=32
 665 | size=1
 666 | stride=1
 667 | pad=1
 668 | activation=leaky
 669 | 
 670 | [convolutional]
 671 | batch_normalize=1
 672 | filters=40
 673 | size=3
 674 | stride=1
 675 | pad=1
 676 | activation=leaky
 677 | 
 678 | [route]
 679 | layers=-1,-9
 680 | 
 681 | [convolutional]
 682 | batch_normalize=1
 683 | filters=64
 684 | size=1
 685 | stride=1
 686 | pad=1
 687 | activation=leaky
 688 | 
 689 | [convolutional]
 690 | batch_normalize=1
 691 | filters=512
 692 | size=1
 693 | stride=1
 694 | pad=1
 695 | activation=leaky
 696 | 
 697 | [upsample]
 698 | stride=2
 699 | 
 700 | [route]
 701 | layers=-1,-23
 702 | 
 703 | [convolutional]
 704 | batch_normalize=1
 705 | filters=64
 706 | size=1
 707 | stride=1
 708 | pad=1
 709 | activation=leaky
 710 | 
 711 | [route]
 712 | layers=-2
 713 | 
 714 | [convolutional]
 715 | batch_normalize=1
 716 | filters=96
 717 | size=1
 718 | stride=1
 719 | pad=1
 720 | activation=leaky
 721 | 
 722 | [convolutional]
 723 | batch_normalize=1
 724 | filters=120
 725 | size=1
 726 | stride=1
 727 | pad=1
 728 | activation=leaky
 729 | 
 730 | [convolutional]
 731 | batch_normalize=1
 732 | filters=152
 733 | size=3
 734 | stride=1
 735 | pad=1
 736 | activation=leaky
 737 | 
 738 | [convolutional]
 739 | batch_normalize=1
 740 | filters=144
 741 | size=1
 742 | stride=1
 743 | pad=1
 744 | activation=leaky
 745 | 
 746 | [convolutional]
 747 | batch_normalize=1
 748 | filters=144
 749 | size=3
 750 | stride=1
 751 | pad=1
 752 | activation=leaky
 753 | 
 754 | [convolutional]
 755 | batch_normalize=1
 756 | filters=112
 757 | size=1
 758 | stride=1
 759 | pad=1
 760 | activation=leaky
 761 | 
 762 | [convolutional]
 763 | batch_normalize=1
 764 | filters=120
 765 | size=3
 766 | stride=1
 767 | pad=1
 768 | activation=leaky
 769 | 
 770 | [route]
 771 | layers=-1,-9
 772 | 
 773 | [convolutional]
 774 | batch_normalize=1
 775 | filters=120
 776 | size=1
 777 | stride=1
 778 | pad=1
 779 | activation=leaky
 780 | 
 781 | [convolutional]
 782 | batch_normalize=1
 783 | filters=256
 784 | size=1
 785 | stride=1
 786 | pad=1
 787 | activation=leaky
 788 | 
 789 | [upsample]
 790 | stride=2
 791 | 
 792 | [route]
 793 | layers=-1,-70
 794 | 
 795 | [convolutional]
 796 | batch_normalize=1
 797 | filters=72
 798 | size=1
 799 | stride=1
 800 | pad=1
 801 | activation=leaky
 802 | 
 803 | [route]
 804 | layers=-2
 805 | 
 806 | [convolutional]
 807 | batch_normalize=1
 808 | filters=80
 809 | size=1
 810 | stride=1
 811 | pad=1
 812 | activation=leaky
 813 | 
 814 | [convolutional]
 815 | batch_normalize=1
 816 | filters=88
 817 | size=1
 818 | stride=1
 819 | pad=1
 820 | activation=leaky
 821 | 
 822 | [convolutional]
 823 | batch_normalize=1
 824 | filters=104
 825 | size=3
 826 | stride=1
 827 | pad=1
 828 | activation=leaky
 829 | 
 830 | [convolutional]
 831 | batch_normalize=1
 832 | filters=96
 833 | size=1
 834 | stride=1
 835 | pad=1
 836 | activation=leaky
 837 | 
 838 | [convolutional]
 839 | batch_normalize=1
 840 | filters=96
 841 | size=3
 842 | stride=1
 843 | pad=1
 844 | activation=leaky
 845 | 
 846 | [convolutional]
 847 | batch_normalize=1
 848 | filters=80
 849 | size=1
 850 | stride=1
 851 | pad=1
 852 | activation=leaky
 853 | 
 854 | [convolutional]
 855 | batch_normalize=1
 856 | filters=112
 857 | size=3
 858 | stride=1
 859 | pad=1
 860 | activation=leaky
 861 | 
 862 | [route]
 863 | layers=-1,-9
 864 | 
 865 | [convolutional]
 866 | batch_normalize=1
 867 | filters=208
 868 | size=1
 869 | stride=1
 870 | pad=1
 871 | activation=leaky
 872 | 
 873 | [convolutional]
 874 | size=1
 875 | stride=1
 876 | pad=1
 877 | filters=24
 878 | activation=linear
 879 | 
 880 | [yolo]
 881 | mask=0,1,2
 882 | anchors=40,39, 51,50, 61,59, 75,69, 62,92, 88,98, 115,77, 93,129, 128,115
 883 | classes=3
 884 | num=9
 885 | jitter=.3
 886 | ignore_thresh=.7
 887 | truth_thresh=1
 888 | scale_x_y=1.2
 889 | iou_thresh=0.213
 890 | cls_normalizer=1.0
 891 | iou_normalizer=0.07
 892 | iou_loss=ciou
 893 | nms_kind=greedynms
 894 | beta_nms=0.6
 895 | 
 896 | [route]
 897 | layers=-3
 898 | 
 899 | [convolutional]
 900 | batch_normalize=1
 901 | filters=120
 902 | size=3
 903 | stride=2
 904 | pad=1
 905 | activation=leaky
 906 | 
 907 | [route]
 908 | layers=-1,-18
 909 | 
 910 | [convolutional]
 911 | batch_normalize=1
 912 | filters=128
 913 | size=1
 914 | stride=1
 915 | pad=1
 916 | activation=leaky
 917 | 
 918 | [route]
 919 | layers=-2
 920 | 
 921 | [convolutional]
 922 | batch_normalize=1
 923 | filters=112
 924 | size=1
 925 | stride=1
 926 | pad=1
 927 | activation=leaky
 928 | 
 929 | [convolutional]
 930 | batch_normalize=1
 931 | filters=104
 932 | size=1
 933 | stride=1
 934 | pad=1
 935 | activation=leaky
 936 | 
 937 | [convolutional]
 938 | batch_normalize=1
 939 | filters=96
 940 | size=3
 941 | stride=1
 942 | pad=1
 943 | activation=leaky
 944 | 
 945 | [convolutional]
 946 | batch_normalize=1
 947 | filters=80
 948 | size=1
 949 | stride=1
 950 | pad=1
 951 | activation=leaky
 952 | 
 953 | [convolutional]
 954 | batch_normalize=1
 955 | filters=96
 956 | size=3
 957 | stride=1
 958 | pad=1
 959 | activation=leaky
 960 | 
 961 | [convolutional]
 962 | batch_normalize=1
 963 | filters=88
 964 | size=1
 965 | stride=1
 966 | pad=1
 967 | activation=leaky
 968 | 
 969 | [convolutional]
 970 | batch_normalize=1
 971 | filters=112
 972 | size=3
 973 | stride=1
 974 | pad=1
 975 | activation=leaky
 976 | 
 977 | [route]
 978 | layers=-1,-9
 979 | 
 980 | [convolutional]
 981 | batch_normalize=1
 982 | filters=240
 983 | size=1
 984 | stride=1
 985 | pad=1
 986 | activation=leaky
 987 | 
 988 | [convolutional]
 989 | size=1
 990 | stride=1
 991 | pad=1
 992 | filters=24
 993 | activation=linear
 994 | 
 995 | [yolo]
 996 | mask=3,4,5
 997 | anchors=40,39, 51,50, 61,59, 75,69, 62,92, 88,98, 115,77, 93,129, 128,115
 998 | classes=3
 999 | num=9
1000 | jitter=.3
1001 | ignore_thresh=.7
1002 | truth_thresh=1
1003 | scale_x_y=1.2
1004 | iou_thresh=0.213
1005 | cls_normalizer=1.0
1006 | iou_normalizer=0.07
1007 | iou_loss=ciou
1008 | nms_kind=greedynms
1009 | beta_nms=0.6
1010 | 
1011 | [route]
1012 | layers=-3
1013 | 
1014 | [convolutional]
1015 | batch_normalize=1
1016 | filters=224
1017 | size=3
1018 | stride=2
1019 | pad=1
1020 | activation=leaky
1021 | 
1022 | [route]
1023 | layers=-1,-48
1024 | 
1025 | [convolutional]
1026 | batch_normalize=1
1027 | filters=104
1028 | size=1
1029 | stride=1
1030 | pad=1
1031 | activation=leaky
1032 | 
1033 | [route]
1034 | layers=-2
1035 | 
1036 | [convolutional]
1037 | batch_normalize=1
1038 | filters=40
1039 | size=1
1040 | stride=1
1041 | pad=1
1042 | activation=leaky
1043 | 
1044 | [convolutional]
1045 | batch_normalize=1
1046 | filters=40
1047 | size=1
1048 | stride=1
1049 | pad=1
1050 | activation=leaky
1051 | 
1052 | [convolutional]
1053 | batch_normalize=1
1054 | filters=40
1055 | size=3
1056 | stride=1
1057 | pad=1
1058 | activation=leaky
1059 | 
1060 | [convolutional]
1061 | batch_normalize=1
1062 | filters=32
1063 | size=1
1064 | stride=1
1065 | pad=1
1066 | activation=leaky
1067 | 
1068 | [convolutional]
1069 | batch_normalize=1
1070 | filters=48
1071 | size=3
1072 | stride=1
1073 | pad=1
1074 | activation=leaky
1075 | 
1076 | [convolutional]
1077 | batch_normalize=1
1078 | filters=48
1079 | size=1
1080 | stride=1
1081 | pad=1
1082 | activation=leaky
1083 | 
1084 | [convolutional]
1085 | batch_normalize=1
1086 | filters=40
1087 | size=3
1088 | stride=1
1089 | pad=1
1090 | activation=leaky
1091 | 
1092 | [route]
1093 | layers=-1,-9
1094 | 
1095 | [convolutional]
1096 | batch_normalize=1
1097 | filters=232
1098 | size=1
1099 | stride=1
1100 | pad=1
1101 | activation=leaky
1102 | 
1103 | [convolutional]
1104 | size=1
1105 | stride=1
1106 | pad=1
1107 | filters=24
1108 | activation=linear
1109 | 
1110 | [yolo]
1111 | mask=6,7,8
1112 | anchors=40,39, 51,50, 61,59, 75,69, 62,92, 88,98, 115,77, 93,129, 128,115
1113 | classes=3
1114 | num=9
1115 | jitter=.3
1116 | ignore_thresh=.7
1117 | truth_thresh=1
1118 | scale_x_y=1.2
1119 | iou_thresh=0.213
1120 | cls_normalizer=1.0
1121 | iou_normalizer=0.07
1122 | iou_loss=ciou
1123 | nms_kind=greedynms
1124 | beta_nms=0.6
1125 | 
1126 | 


--------------------------------------------------------------------------------
/cfg/0514/prune_0.8_keep_0.01_8x_yolov5l_v4.cfg:
--------------------------------------------------------------------------------
   1 | [net]
   2 | batch=64
   3 | subdivisions=8
   4 | width=416
   5 | height=416
   6 | channels=3
   7 | momentum=0.949
   8 | decay=0.0005
   9 | angle=0
  10 | saturation=1.5
  11 | exposure=1.5
  12 | hue=.1
  13 | learning_rate=0.00261
  14 | burn_in=1000
  15 | max_batches=500500
  16 | policy=steps
  17 | steps=400000,450000
  18 | scales=.1,.1
  19 | mosaic=1
  20 | 
  21 | [focus]
  22 | filters=12
  23 | 
  24 | [convolutional]
  25 | batch_normalize=1
  26 | filters=40
  27 | size=3
  28 | stride=1
  29 | pad=1
  30 | activation=leaky
  31 | 
  32 | [convolutional]
  33 | batch_normalize=1
  34 | filters=112
  35 | size=3
  36 | stride=2
  37 | pad=1
  38 | activation=leaky
  39 | 
  40 | [convolutional]
  41 | batch_normalize=1
  42 | filters=64
  43 | size=1
  44 | stride=1
  45 | pad=1
  46 | activation=leaky
  47 | 
  48 | [route]
  49 | layers=-2
  50 | 
  51 | [convolutional]
  52 | batch_normalize=1
  53 | filters=64
  54 | size=1
  55 | stride=1
  56 | pad=1
  57 | activation=leaky
  58 | 
  59 | [convolutional]
  60 | batch_normalize=1
  61 | filters=64
  62 | size=1
  63 | stride=1
  64 | pad=1
  65 | activation=leaky
  66 | 
  67 | [convolutional]
  68 | batch_normalize=1
  69 | filters=64
  70 | size=3
  71 | stride=1
  72 | pad=1
  73 | activation=leaky
  74 | 
  75 | [shortcut]
  76 | from=-3
  77 | activation=linear
  78 | 
  79 | [convolutional]
  80 | batch_normalize=1
  81 | filters=64
  82 | size=1
  83 | stride=1
  84 | pad=1
  85 | activation=leaky
  86 | 
  87 | [convolutional]
  88 | batch_normalize=1
  89 | filters=64
  90 | size=3
  91 | stride=1
  92 | pad=1
  93 | activation=leaky
  94 | 
  95 | [shortcut]
  96 | from=-3
  97 | activation=linear
  98 | 
  99 | [convolutional]
 100 | batch_normalize=1
 101 | filters=64
 102 | size=1
 103 | stride=1
 104 | pad=1
 105 | activation=leaky
 106 | 
 107 | [convolutional]
 108 | batch_normalize=1
 109 | filters=64
 110 | size=3
 111 | stride=1
 112 | pad=1
 113 | activation=leaky
 114 | 
 115 | [shortcut]
 116 | from=-3
 117 | activation=linear
 118 | 
 119 | [route]
 120 | layers=-1,-12
 121 | 
 122 | [convolutional]
 123 | batch_normalize=1
 124 | filters=128
 125 | size=1
 126 | stride=1
 127 | pad=1
 128 | activation=leaky
 129 | 
 130 | [convolutional]
 131 | batch_normalize=1
 132 | filters=248
 133 | size=3
 134 | stride=2
 135 | pad=1
 136 | activation=leaky
 137 | 
 138 | [convolutional]
 139 | batch_normalize=1
 140 | filters=128
 141 | size=1
 142 | stride=1
 143 | pad=1
 144 | activation=leaky
 145 | 
 146 | [route]
 147 | layers=-2
 148 | 
 149 | [convolutional]
 150 | batch_normalize=1
 151 | filters=128
 152 | size=1
 153 | stride=1
 154 | pad=1
 155 | activation=leaky
 156 | 
 157 | [convolutional]
 158 | batch_normalize=1
 159 | filters=56
 160 | size=1
 161 | stride=1
 162 | pad=1
 163 | activation=leaky
 164 | 
 165 | [convolutional]
 166 | batch_normalize=1
 167 | filters=128
 168 | size=3
 169 | stride=1
 170 | pad=1
 171 | activation=leaky
 172 | 
 173 | [shortcut]
 174 | from=-3
 175 | activation=linear
 176 | 
 177 | [convolutional]
 178 | batch_normalize=1
 179 | filters=56
 180 | size=1
 181 | stride=1
 182 | pad=1
 183 | activation=leaky
 184 | 
 185 | [convolutional]
 186 | batch_normalize=1
 187 | filters=128
 188 | size=3
 189 | stride=1
 190 | pad=1
 191 | activation=leaky
 192 | 
 193 | [shortcut]
 194 | from=-3
 195 | activation=linear
 196 | 
 197 | [convolutional]
 198 | batch_normalize=1
 199 | filters=24
 200 | size=1
 201 | stride=1
 202 | pad=1
 203 | activation=leaky
 204 | 
 205 | [convolutional]
 206 | batch_normalize=1
 207 | filters=128
 208 | size=3
 209 | stride=1
 210 | pad=1
 211 | activation=leaky
 212 | 
 213 | [shortcut]
 214 | from=-3
 215 | activation=linear
 216 | 
 217 | [convolutional]
 218 | batch_normalize=1
 219 | filters=24
 220 | size=1
 221 | stride=1
 222 | pad=1
 223 | activation=leaky
 224 | 
 225 | [convolutional]
 226 | batch_normalize=1
 227 | filters=128
 228 | size=3
 229 | stride=1
 230 | pad=1
 231 | activation=leaky
 232 | 
 233 | [shortcut]
 234 | from=-3
 235 | activation=linear
 236 | 
 237 | [convolutional]
 238 | batch_normalize=1
 239 | filters=40
 240 | size=1
 241 | stride=1
 242 | pad=1
 243 | activation=leaky
 244 | 
 245 | [convolutional]
 246 | batch_normalize=1
 247 | filters=128
 248 | size=3
 249 | stride=1
 250 | pad=1
 251 | activation=leaky
 252 | 
 253 | [shortcut]
 254 | from=-3
 255 | activation=linear
 256 | 
 257 | [convolutional]
 258 | batch_normalize=1
 259 | filters=32
 260 | size=1
 261 | stride=1
 262 | pad=1
 263 | activation=leaky
 264 | 
 265 | [convolutional]
 266 | batch_normalize=1
 267 | filters=128
 268 | size=3
 269 | stride=1
 270 | pad=1
 271 | activation=leaky
 272 | 
 273 | [shortcut]
 274 | from=-3
 275 | activation=linear
 276 | 
 277 | [convolutional]
 278 | batch_normalize=1
 279 | filters=64
 280 | size=1
 281 | stride=1
 282 | pad=1
 283 | activation=leaky
 284 | 
 285 | [convolutional]
 286 | batch_normalize=1
 287 | filters=128
 288 | size=3
 289 | stride=1
 290 | pad=1
 291 | activation=leaky
 292 | 
 293 | [shortcut]
 294 | from=-3
 295 | activation=linear
 296 | 
 297 | [convolutional]
 298 | batch_normalize=1
 299 | filters=64
 300 | size=1
 301 | stride=1
 302 | pad=1
 303 | activation=leaky
 304 | 
 305 | [convolutional]
 306 | batch_normalize=1
 307 | filters=128
 308 | size=3
 309 | stride=1
 310 | pad=1
 311 | activation=leaky
 312 | 
 313 | [shortcut]
 314 | from=-3
 315 | activation=linear
 316 | 
 317 | [convolutional]
 318 | batch_normalize=1
 319 | filters=56
 320 | size=1
 321 | stride=1
 322 | pad=1
 323 | activation=leaky
 324 | 
 325 | [convolutional]
 326 | batch_normalize=1
 327 | filters=128
 328 | size=3
 329 | stride=1
 330 | pad=1
 331 | activation=leaky
 332 | 
 333 | [shortcut]
 334 | from=-3
 335 | activation=linear
 336 | 
 337 | [route]
 338 | layers=-1,-30
 339 | 
 340 | [convolutional]
 341 | batch_normalize=1
 342 | filters=240
 343 | size=1
 344 | stride=1
 345 | pad=1
 346 | activation=leaky
 347 | 
 348 | [convolutional]
 349 | batch_normalize=1
 350 | filters=240
 351 | size=3
 352 | stride=2
 353 | pad=1
 354 | activation=leaky
 355 | 
 356 | [convolutional]
 357 | batch_normalize=1
 358 | filters=96
 359 | size=1
 360 | stride=1
 361 | pad=1
 362 | activation=leaky
 363 | 
 364 | [route]
 365 | layers=-2
 366 | 
 367 | [convolutional]
 368 | batch_normalize=1
 369 | filters=245
 370 | size=1
 371 | stride=1
 372 | pad=1
 373 | activation=leaky
 374 | 
 375 | [convolutional]
 376 | batch_normalize=1
 377 | filters=56
 378 | size=1
 379 | stride=1
 380 | pad=1
 381 | activation=leaky
 382 | 
 383 | [convolutional]
 384 | batch_normalize=1
 385 | filters=245
 386 | size=3
 387 | stride=1
 388 | pad=1
 389 | activation=leaky
 390 | 
 391 | [shortcut]
 392 | from=-3
 393 | activation=linear
 394 | 
 395 | [convolutional]
 396 | batch_normalize=1
 397 | filters=8
 398 | size=1
 399 | stride=1
 400 | pad=1
 401 | activation=leaky
 402 | 
 403 | [convolutional]
 404 | batch_normalize=1
 405 | filters=245
 406 | size=3
 407 | stride=1
 408 | pad=1
 409 | activation=leaky
 410 | 
 411 | [shortcut]
 412 | from=-3
 413 | activation=linear
 414 | 
 415 | [convolutional]
 416 | batch_normalize=1
 417 | filters=16
 418 | size=1
 419 | stride=1
 420 | pad=1
 421 | activation=leaky
 422 | 
 423 | [convolutional]
 424 | batch_normalize=1
 425 | filters=245
 426 | size=3
 427 | stride=1
 428 | pad=1
 429 | activation=leaky
 430 | 
 431 | [shortcut]
 432 | from=-3
 433 | activation=linear
 434 | 
 435 | [convolutional]
 436 | batch_normalize=1
 437 | filters=24
 438 | size=1
 439 | stride=1
 440 | pad=1
 441 | activation=leaky
 442 | 
 443 | [convolutional]
 444 | batch_normalize=1
 445 | filters=245
 446 | size=3
 447 | stride=1
 448 | pad=1
 449 | activation=leaky
 450 | 
 451 | [shortcut]
 452 | from=-3
 453 | activation=linear
 454 | 
 455 | [convolutional]
 456 | batch_normalize=1
 457 | filters=16
 458 | size=1
 459 | stride=1
 460 | pad=1
 461 | activation=leaky
 462 | 
 463 | [convolutional]
 464 | batch_normalize=1
 465 | filters=245
 466 | size=3
 467 | stride=1
 468 | pad=1
 469 | activation=leaky
 470 | 
 471 | [shortcut]
 472 | from=-3
 473 | activation=linear
 474 | 
 475 | [convolutional]
 476 | batch_normalize=1
 477 | filters=24
 478 | size=1
 479 | stride=1
 480 | pad=1
 481 | activation=leaky
 482 | 
 483 | [convolutional]
 484 | batch_normalize=1
 485 | filters=245
 486 | size=3
 487 | stride=1
 488 | pad=1
 489 | activation=leaky
 490 | 
 491 | [shortcut]
 492 | from=-3
 493 | activation=linear
 494 | 
 495 | [convolutional]
 496 | batch_normalize=1
 497 | filters=16
 498 | size=1
 499 | stride=1
 500 | pad=1
 501 | activation=leaky
 502 | 
 503 | [convolutional]
 504 | batch_normalize=1
 505 | filters=245
 506 | size=3
 507 | stride=1
 508 | pad=1
 509 | activation=leaky
 510 | 
 511 | [shortcut]
 512 | from=-3
 513 | activation=linear
 514 | 
 515 | [convolutional]
 516 | batch_normalize=1
 517 | filters=32
 518 | size=1
 519 | stride=1
 520 | pad=1
 521 | activation=leaky
 522 | 
 523 | [convolutional]
 524 | batch_normalize=1
 525 | filters=245
 526 | size=3
 527 | stride=1
 528 | pad=1
 529 | activation=leaky
 530 | 
 531 | [shortcut]
 532 | from=-3
 533 | activation=linear
 534 | 
 535 | [convolutional]
 536 | batch_normalize=1
 537 | filters=24
 538 | size=1
 539 | stride=1
 540 | pad=1
 541 | activation=leaky
 542 | 
 543 | [convolutional]
 544 | batch_normalize=1
 545 | filters=245
 546 | size=3
 547 | stride=1
 548 | pad=1
 549 | activation=leaky
 550 | 
 551 | [shortcut]
 552 | from=-3
 553 | activation=linear
 554 | 
 555 | [route]
 556 | layers=-1,-30
 557 | 
 558 | [convolutional]
 559 | batch_normalize=1
 560 | filters=144
 561 | size=1
 562 | stride=1
 563 | pad=1
 564 | activation=leaky
 565 | 
 566 | [convolutional]
 567 | batch_normalize=1
 568 | filters=104
 569 | size=3
 570 | stride=2
 571 | pad=1
 572 | activation=leaky
 573 | 
 574 | [convolutional]
 575 | batch_normalize=1
 576 | filters=512
 577 | size=1
 578 | stride=1
 579 | pad=1
 580 | activation=leaky
 581 | 
 582 | [maxpool]
 583 | stride=1
 584 | size=5
 585 | 
 586 | [route]
 587 | layers=-2
 588 | 
 589 | [maxpool]
 590 | stride=1
 591 | size=9
 592 | 
 593 | [route]
 594 | layers=-4
 595 | 
 596 | [maxpool]
 597 | stride=1
 598 | size=13
 599 | 
 600 | [route]
 601 | layers=-6,-5,-3,-1
 602 | 
 603 | [convolutional]
 604 | batch_normalize=1
 605 | filters=56
 606 | size=1
 607 | stride=1
 608 | pad=1
 609 | activation=leaky
 610 | 
 611 | [convolutional]
 612 | batch_normalize=1
 613 | filters=40
 614 | size=1
 615 | stride=1
 616 | pad=1
 617 | activation=leaky
 618 | 
 619 | [route]
 620 | layers=-2
 621 | 
 622 | [convolutional]
 623 | batch_normalize=1
 624 | filters=8
 625 | size=1
 626 | stride=1
 627 | pad=1
 628 | activation=leaky
 629 | 
 630 | [convolutional]
 631 | batch_normalize=1
 632 | filters=8
 633 | size=1
 634 | stride=1
 635 | pad=1
 636 | activation=leaky
 637 | 
 638 | [convolutional]
 639 | batch_normalize=1
 640 | filters=8
 641 | size=3
 642 | stride=1
 643 | pad=1
 644 | activation=leaky
 645 | 
 646 | [convolutional]
 647 | batch_normalize=1
 648 | filters=8
 649 | size=1
 650 | stride=1
 651 | pad=1
 652 | activation=leaky
 653 | 
 654 | [convolutional]
 655 | batch_normalize=1
 656 | filters=16
 657 | size=3
 658 | stride=1
 659 | pad=1
 660 | activation=leaky
 661 | 
 662 | [convolutional]
 663 | batch_normalize=1
 664 | filters=8
 665 | size=1
 666 | stride=1
 667 | pad=1
 668 | activation=leaky
 669 | 
 670 | [convolutional]
 671 | batch_normalize=1
 672 | filters=8
 673 | size=3
 674 | stride=1
 675 | pad=1
 676 | activation=leaky
 677 | 
 678 | [route]
 679 | layers=-1,-9
 680 | 
 681 | [convolutional]
 682 | batch_normalize=1
 683 | filters=40
 684 | size=1
 685 | stride=1
 686 | pad=1
 687 | activation=leaky
 688 | 
 689 | [convolutional]
 690 | batch_normalize=1
 691 | filters=512
 692 | size=1
 693 | stride=1
 694 | pad=1
 695 | activation=leaky
 696 | 
 697 | [upsample]
 698 | stride=2
 699 | 
 700 | [route]
 701 | layers=-1,-23
 702 | 
 703 | [convolutional]
 704 | batch_normalize=1
 705 | filters=48
 706 | size=1
 707 | stride=1
 708 | pad=1
 709 | activation=leaky
 710 | 
 711 | [route]
 712 | layers=-2
 713 | 
 714 | [convolutional]
 715 | batch_normalize=1
 716 | filters=80
 717 | size=1
 718 | stride=1
 719 | pad=1
 720 | activation=leaky
 721 | 
 722 | [convolutional]
 723 | batch_normalize=1
 724 | filters=104
 725 | size=1
 726 | stride=1
 727 | pad=1
 728 | activation=leaky
 729 | 
 730 | [convolutional]
 731 | batch_normalize=1
 732 | filters=128
 733 | size=3
 734 | stride=1
 735 | pad=1
 736 | activation=leaky
 737 | 
 738 | [convolutional]
 739 | batch_normalize=1
 740 | filters=112
 741 | size=1
 742 | stride=1
 743 | pad=1
 744 | activation=leaky
 745 | 
 746 | [convolutional]
 747 | batch_normalize=1
 748 | filters=112
 749 | size=3
 750 | stride=1
 751 | pad=1
 752 | activation=leaky
 753 | 
 754 | [convolutional]
 755 | batch_normalize=1
 756 | filters=96
 757 | size=1
 758 | stride=1
 759 | pad=1
 760 | activation=leaky
 761 | 
 762 | [convolutional]
 763 | batch_normalize=1
 764 | filters=104
 765 | size=3
 766 | stride=1
 767 | pad=1
 768 | activation=leaky
 769 | 
 770 | [route]
 771 | layers=-1,-9
 772 | 
 773 | [convolutional]
 774 | batch_normalize=1
 775 | filters=88
 776 | size=1
 777 | stride=1
 778 | pad=1
 779 | activation=leaky
 780 | 
 781 | [convolutional]
 782 | batch_normalize=1
 783 | filters=256
 784 | size=1
 785 | stride=1
 786 | pad=1
 787 | activation=leaky
 788 | 
 789 | [upsample]
 790 | stride=2
 791 | 
 792 | [route]
 793 | layers=-1,-70
 794 | 
 795 | [convolutional]
 796 | batch_normalize=1
 797 | filters=32
 798 | size=1
 799 | stride=1
 800 | pad=1
 801 | activation=leaky
 802 | 
 803 | [route]
 804 | layers=-2
 805 | 
 806 | [convolutional]
 807 | batch_normalize=1
 808 | filters=72
 809 | size=1
 810 | stride=1
 811 | pad=1
 812 | activation=leaky
 813 | 
 814 | [convolutional]
 815 | batch_normalize=1
 816 | filters=72
 817 | size=1
 818 | stride=1
 819 | pad=1
 820 | activation=leaky
 821 | 
 822 | [convolutional]
 823 | batch_normalize=1
 824 | filters=96
 825 | size=3
 826 | stride=1
 827 | pad=1
 828 | activation=leaky
 829 | 
 830 | [convolutional]
 831 | batch_normalize=1
 832 | filters=88
 833 | size=1
 834 | stride=1
 835 | pad=1
 836 | activation=leaky
 837 | 
 838 | [convolutional]
 839 | batch_normalize=1
 840 | filters=88
 841 | size=3
 842 | stride=1
 843 | pad=1
 844 | activation=leaky
 845 | 
 846 | [convolutional]
 847 | batch_normalize=1
 848 | filters=80
 849 | size=1
 850 | stride=1
 851 | pad=1
 852 | activation=leaky
 853 | 
 854 | [convolutional]
 855 | batch_normalize=1
 856 | filters=96
 857 | size=3
 858 | stride=1
 859 | pad=1
 860 | activation=leaky
 861 | 
 862 | [route]
 863 | layers=-1,-9
 864 | 
 865 | [convolutional]
 866 | batch_normalize=1
 867 | filters=96
 868 | size=1
 869 | stride=1
 870 | pad=1
 871 | activation=leaky
 872 | 
 873 | [convolutional]
 874 | size=1
 875 | stride=1
 876 | pad=1
 877 | filters=24
 878 | activation=linear
 879 | 
 880 | [yolo]
 881 | mask=0,1,2
 882 | anchors=39,38, 49,67, 74,49, 74,86, 113,71, 97,119, 163,108, 134,155, 210,199
 883 | classes=3
 884 | num=9
 885 | jitter=.3
 886 | ignore_thresh=.7
 887 | truth_thresh=1
 888 | scale_x_y=1.2
 889 | iou_thresh=0.213
 890 | cls_normalizer=1.0
 891 | iou_normalizer=0.07
 892 | iou_loss=ciou
 893 | nms_kind=greedynms
 894 | beta_nms=0.6
 895 | 
 896 | [route]
 897 | layers=-3
 898 | 
 899 | [convolutional]
 900 | batch_normalize=1
 901 | filters=72
 902 | size=3
 903 | stride=2
 904 | pad=1
 905 | activation=leaky
 906 | 
 907 | [route]
 908 | layers=-1,-18
 909 | 
 910 | [convolutional]
 911 | batch_normalize=1
 912 | filters=64
 913 | size=1
 914 | stride=1
 915 | pad=1
 916 | activation=leaky
 917 | 
 918 | [route]
 919 | layers=-2
 920 | 
 921 | [convolutional]
 922 | batch_normalize=1
 923 | filters=56
 924 | size=1
 925 | stride=1
 926 | pad=1
 927 | activation=leaky
 928 | 
 929 | [convolutional]
 930 | batch_normalize=1
 931 | filters=48
 932 | size=1
 933 | stride=1
 934 | pad=1
 935 | activation=leaky
 936 | 
 937 | [convolutional]
 938 | batch_normalize=1
 939 | filters=64
 940 | size=3
 941 | stride=1
 942 | pad=1
 943 | activation=leaky
 944 | 
 945 | [convolutional]
 946 | batch_normalize=1
 947 | filters=48
 948 | size=1
 949 | stride=1
 950 | pad=1
 951 | activation=leaky
 952 | 
 953 | [convolutional]
 954 | batch_normalize=1
 955 | filters=64
 956 | size=3
 957 | stride=1
 958 | pad=1
 959 | activation=leaky
 960 | 
 961 | [convolutional]
 962 | batch_normalize=1
 963 | filters=64
 964 | size=1
 965 | stride=1
 966 | pad=1
 967 | activation=leaky
 968 | 
 969 | [convolutional]
 970 | batch_normalize=1
 971 | filters=80
 972 | size=3
 973 | stride=1
 974 | pad=1
 975 | activation=leaky
 976 | 
 977 | [route]
 978 | layers=-1,-9
 979 | 
 980 | [convolutional]
 981 | batch_normalize=1
 982 | filters=104
 983 | size=1
 984 | stride=1
 985 | pad=1
 986 | activation=leaky
 987 | 
 988 | [convolutional]
 989 | size=1
 990 | stride=1
 991 | pad=1
 992 | filters=24
 993 | activation=linear
 994 | 
 995 | [yolo]
 996 | mask=3,4,5
 997 | anchors=39,38, 49,67, 74,49, 74,86, 113,71, 97,119, 163,108, 134,155, 210,199
 998 | classes=3
 999 | num=9
1000 | jitter=.3
1001 | ignore_thresh=.7
1002 | truth_thresh=1
1003 | scale_x_y=1.2
1004 | iou_thresh=0.213
1005 | cls_normalizer=1.0
1006 | iou_normalizer=0.07
1007 | iou_loss=ciou
1008 | nms_kind=greedynms
1009 | beta_nms=0.6
1010 | 
1011 | [route]
1012 | layers=-3
1013 | 
1014 | [convolutional]
1015 | batch_normalize=1
1016 | filters=136
1017 | size=3
1018 | stride=2
1019 | pad=1
1020 | activation=leaky
1021 | 
1022 | [route]
1023 | layers=-1,-48
1024 | 
1025 | [convolutional]
1026 | batch_normalize=1
1027 | filters=64
1028 | size=1
1029 | stride=1
1030 | pad=1
1031 | activation=leaky
1032 | 
1033 | [route]
1034 | layers=-2
1035 | 
1036 | [convolutional]
1037 | batch_normalize=1
1038 | filters=8
1039 | size=1
1040 | stride=1
1041 | pad=1
1042 | activation=leaky
1043 | 
1044 | [convolutional]
1045 | batch_normalize=1
1046 | filters=8
1047 | size=1
1048 | stride=1
1049 | pad=1
1050 | activation=leaky
1051 | 
1052 | [convolutional]
1053 | batch_normalize=1
1054 | filters=8
1055 | size=3
1056 | stride=1
1057 | pad=1
1058 | activation=leaky
1059 | 
1060 | [convolutional]
1061 | batch_normalize=1
1062 | filters=8
1063 | size=1
1064 | stride=1
1065 | pad=1
1066 | activation=leaky
1067 | 
1068 | [convolutional]
1069 | batch_normalize=1
1070 | filters=8
1071 | size=3
1072 | stride=1
1073 | pad=1
1074 | activation=leaky
1075 | 
1076 | [convolutional]
1077 | batch_normalize=1
1078 | filters=8
1079 | size=1
1080 | stride=1
1081 | pad=1
1082 | activation=leaky
1083 | 
1084 | [convolutional]
1085 | batch_normalize=1
1086 | filters=8
1087 | size=3
1088 | stride=1
1089 | pad=1
1090 | activation=leaky
1091 | 
1092 | [route]
1093 | layers=-1,-9
1094 | 
1095 | [convolutional]
1096 | batch_normalize=1
1097 | filters=72
1098 | size=1
1099 | stride=1
1100 | pad=1
1101 | activation=leaky
1102 | 
1103 | [convolutional]
1104 | size=1
1105 | stride=1
1106 | pad=1
1107 | filters=24
1108 | activation=linear
1109 | 
1110 | [yolo]
1111 | mask=6,7,8
1112 | anchors=39,38, 49,67, 74,49, 74,86, 113,71, 97,119, 163,108, 134,155, 210,199
1113 | classes=3
1114 | num=9
1115 | jitter=.3
1116 | ignore_thresh=.7
1117 | truth_thresh=1
1118 | scale_x_y=1.2
1119 | iou_thresh=0.213
1120 | cls_normalizer=1.0
1121 | iou_normalizer=0.07
1122 | iou_loss=ciou
1123 | nms_kind=greedynms
1124 | beta_nms=0.6
1125 | 
1126 | 


--------------------------------------------------------------------------------
/cfg/0514/yolov5s_v4.cfg:
--------------------------------------------------------------------------------
  1 | [net]
  2 | # Testing
  3 | #batch=1
  4 | #subdivisions=1
  5 | # Training
  6 | batch=64
  7 | subdivisions=8
  8 | width=608
  9 | height=608
 10 | channels=3
 11 | momentum=0.949
 12 | decay=0.0005
 13 | angle=0
 14 | saturation = 1.5
 15 | exposure = 1.5
 16 | hue=.1
 17 | 
 18 | learning_rate=0.00261
 19 | burn_in=1000
 20 | max_batches = 500500
 21 | policy=steps
 22 | steps=400000,450000
 23 | scales=.1,.1
 24 | 
 25 | #cutmix=1
 26 | mosaic=1
 27 | 
 28 | #:104x104 54:52x52 85:26x26 104:13x13 for 416
 29 | [focus]
 30 | filters=12
 31 | 
 32 | [convolutional]
 33 | batch_normalize=1
 34 | filters=32
 35 | size=3
 36 | stride=1
 37 | pad=1
 38 | activation=SiLU
 39 | 
 40 | # Downsample
 41 | [convolutional]
 42 | batch_normalize=1
 43 | filters=64
 44 | size=3
 45 | stride=2
 46 | pad=1
 47 | activation=SiLU
 48 | 
 49 | #C3
 50 | [convolutional]
 51 | batch_normalize=1
 52 | filters=32
 53 | size=1
 54 | stride=1
 55 | pad=1
 56 | activation=SiLU
 57 | 
 58 | [route]
 59 | layers = -2
 60 | 
 61 | [convolutional]
 62 | batch_normalize=1
 63 | filters=32
 64 | size=1
 65 | stride=1
 66 | pad=1
 67 | activation=SiLU
 68 | 
 69 | [convolutional]
 70 | batch_normalize=1
 71 | filters=32
 72 | size=1
 73 | stride=1
 74 | pad=1
 75 | activation=SiLU
 76 | 
 77 | [convolutional]
 78 | batch_normalize=1
 79 | filters=32
 80 | size=3
 81 | stride=1
 82 | pad=1
 83 | activation=SiLU
 84 | 
 85 | [shortcut]
 86 | from=-3
 87 | activation=linear
 88 | 
 89 | [route]
 90 | layers = -1,-6
 91 | 
 92 | [convolutional]
 93 | batch_normalize=1
 94 | filters=64
 95 | size=1
 96 | stride=1
 97 | pad=1
 98 | activation=SiLU
 99 | 
100 | # Downsample
101 | [convolutional]
102 | batch_normalize=1
103 | filters=128
104 | size=3
105 | stride=2
106 | pad=1
107 | activation=SiLU
108 | 
109 | #C3
110 | [convolutional]
111 | batch_normalize=1
112 | filters=64
113 | size=1
114 | stride=1
115 | pad=1
116 | activation=SiLU
117 | 
118 | [route]
119 | layers = -2
120 | 
121 | [convolutional]
122 | batch_normalize=1
123 | filters=64
124 | size=1
125 | stride=1
126 | pad=1
127 | activation=SiLU
128 | 
129 | [convolutional]
130 | batch_normalize=1
131 | filters=64
132 | size=1
133 | stride=1
134 | pad=1
135 | activation=SiLU
136 | 
137 | [convolutional]
138 | batch_normalize=1
139 | filters=64
140 | size=3
141 | stride=1
142 | pad=1
143 | activation=SiLU
144 | 
145 | [shortcut]
146 | from=-3
147 | activation=linear
148 | 
149 | [convolutional]
150 | batch_normalize=1
151 | filters=64
152 | size=1
153 | stride=1
154 | pad=1
155 | activation=SiLU
156 | 
157 | [convolutional]
158 | batch_normalize=1
159 | filters=64
160 | size=3
161 | stride=1
162 | pad=1
163 | activation=SiLU
164 | 
165 | [shortcut]
166 | from=-3
167 | activation=linear
168 | 
169 | [convolutional]
170 | batch_normalize=1
171 | filters=64
172 | size=1
173 | stride=1
174 | pad=1
175 | activation=SiLU
176 | 
177 | [convolutional]
178 | batch_normalize=1
179 | filters=64
180 | size=3
181 | stride=1
182 | pad=1
183 | activation=SiLU
184 | 
185 | [shortcut]
186 | from=-3
187 | activation=linear
188 | 
189 | [route]
190 | layers = -1,-12
191 | 
192 | [convolutional]
193 | batch_normalize=1
194 | filters=128
195 | size=1
196 | stride=1
197 | pad=1
198 | activation=SiLU
199 | 
200 | # Downsample
201 | [convolutional]
202 | batch_normalize=1
203 | filters=256
204 | size=3
205 | stride=2
206 | pad=1
207 | activation=SiLU
208 | 
209 | #C3
210 | [convolutional]
211 | batch_normalize=1
212 | filters=128
213 | size=1
214 | stride=1
215 | pad=1
216 | activation=SiLU
217 | 
218 | [route]
219 | layers = -2
220 | 
221 | [convolutional]
222 | batch_normalize=1
223 | filters=128
224 | size=1
225 | stride=1
226 | pad=1
227 | activation=SiLU
228 | 
229 | [convolutional]
230 | batch_normalize=1
231 | filters=128
232 | size=1
233 | stride=1
234 | pad=1
235 | activation=SiLU
236 | 
237 | [convolutional]
238 | batch_normalize=1
239 | filters=128
240 | size=3
241 | stride=1
242 | pad=1
243 | activation=SiLU
244 | 
245 | [shortcut]
246 | from=-3
247 | activation=linear
248 | 
249 | [convolutional]
250 | batch_normalize=1
251 | filters=128
252 | size=1
253 | stride=1
254 | pad=1
255 | activation=SiLU
256 | 
257 | [convolutional]
258 | batch_normalize=1
259 | filters=128
260 | size=3
261 | stride=1
262 | pad=1
263 | activation=SiLU
264 | 
265 | [shortcut]
266 | from=-3
267 | activation=linear
268 | 
269 | [convolutional]
270 | batch_normalize=1
271 | filters=128
272 | size=1
273 | stride=1
274 | pad=1
275 | activation=SiLU
276 | 
277 | [convolutional]
278 | batch_normalize=1
279 | filters=128
280 | size=3
281 | stride=1
282 | pad=1
283 | activation=SiLU
284 | 
285 | [shortcut]
286 | from=-3
287 | activation=linear
288 | 
289 | [route]
290 | layers = -1,-12
291 | 
292 | [convolutional]
293 | batch_normalize=1
294 | filters=256
295 | size=1
296 | stride=1
297 | pad=1
298 | activation=SiLU
299 | 
300 | # Downsample
301 | [convolutional]
302 | batch_normalize=1
303 | filters=512
304 | size=3
305 | stride=2
306 | pad=1
307 | activation=SiLU
308 | 
309 | [convolutional]
310 | batch_normalize=1
311 | filters=256
312 | size=1
313 | stride=1
314 | pad=1
315 | activation=SiLU
316 | 
317 | ### SPP ###
318 | [maxpool]
319 | stride=1
320 | size=5
321 | 
322 | [route]
323 | layers=-2
324 | 
325 | [maxpool]
326 | stride=1
327 | size=9
328 | 
329 | [route]
330 | layers=-4
331 | 
332 | [maxpool]
333 | stride=1
334 | size=13
335 | 
336 | [route]
337 | ###layers=-1,-3,-5,-6
338 | layers=-6,-5,-3,-1
339 | ### End SPP ###
340 | 
341 | [convolutional]
342 | batch_normalize=1
343 | filters=512
344 | size=1
345 | stride=1
346 | pad=1
347 | activation=SiLU
348 | 
349 | #C3
350 | [convolutional]
351 | batch_normalize=1
352 | filters=256
353 | size=1
354 | stride=1
355 | pad=1
356 | activation=SiLU
357 | 
358 | [route]
359 | layers = -2
360 | 
361 | [convolutional]
362 | batch_normalize=1
363 | filters=256
364 | size=1
365 | stride=1
366 | pad=1
367 | activation=SiLU
368 | 
369 | [convolutional]
370 | batch_normalize=1
371 | filters=256
372 | size=1
373 | stride=1
374 | pad=1
375 | activation=SiLU
376 | 
377 | [convolutional]
378 | batch_normalize=1
379 | filters=256
380 | size=3
381 | stride=1
382 | pad=1
383 | activation=SiLU
384 | 
385 | [route]
386 | layers = -1,-5
387 | 
388 | [convolutional]
389 | batch_normalize=1
390 | filters=512
391 | size=1
392 | stride=1
393 | pad=1
394 | activation=SiLU
395 | 
396 | [convolutional]
397 | batch_normalize=1
398 | filters=256
399 | size=1
400 | stride=1
401 | pad=1
402 | activation=SiLU
403 | 
404 | [upsample]
405 | stride=2
406 | 
407 | [route]
408 | layers = -1,-19
409 | 
410 | #C3
411 | [convolutional]
412 | batch_normalize=1
413 | filters=128
414 | size=1
415 | stride=1
416 | pad=1
417 | activation=SiLU
418 | 
419 | [route]
420 | layers = -2
421 | 
422 | [convolutional]
423 | batch_normalize=1
424 | filters=128
425 | size=1
426 | stride=1
427 | pad=1
428 | activation=SiLU
429 | 
430 | [convolutional]
431 | batch_normalize=1
432 | filters=128
433 | size=1
434 | stride=1
435 | pad=1
436 | activation=SiLU
437 | 
438 | [convolutional]
439 | batch_normalize=1
440 | filters=128
441 | size=3
442 | stride=1
443 | pad=1
444 | activation=SiLU
445 | 
446 | [route]
447 | layers = -1,-5
448 | 
449 | [convolutional]
450 | batch_normalize=1
451 | filters=256
452 | size=1
453 | stride=1
454 | pad=1
455 | activation=SiLU
456 | 
457 | [convolutional]
458 | batch_normalize=1
459 | filters=128
460 | size=1
461 | stride=1
462 | pad=1
463 | activation=SiLU
464 | 
465 | [upsample]
466 | stride=2
467 | 
468 | [route]
469 | layers = -1,-44
470 | 
471 | #C3
472 | [convolutional]
473 | batch_normalize=1
474 | filters=64
475 | size=1
476 | stride=1
477 | pad=1
478 | activation=SiLU
479 | 
480 | [route]
481 | layers = -2
482 | 
483 | [convolutional]
484 | batch_normalize=1
485 | filters=64
486 | size=1
487 | stride=1
488 | pad=1
489 | activation=SiLU
490 | 
491 | [convolutional]
492 | batch_normalize=1
493 | filters=64
494 | size=1
495 | stride=1
496 | pad=1
497 | activation=SiLU
498 | 
499 | [convolutional]
500 | batch_normalize=1
501 | filters=64
502 | size=3
503 | stride=1
504 | pad=1
505 | activation=SiLU
506 | 
507 | [route]
508 | layers = -1,-5
509 | 
510 | [convolutional]
511 | batch_normalize=1
512 | filters=128
513 | size=1
514 | stride=1
515 | pad=1
516 | activation=SiLU
517 | 
518 | ######################
519 | [convolutional]
520 | size=1
521 | stride=1
522 | pad=1
523 | filters=255
524 | activation=linear
525 | 
526 | [yolo]
527 | mask = 0,1,2
528 | anchors = 10,13, 16,30, 33,23, 30,61, 62,45, 59,119, 116,90, 156,198, 373,326
529 | classes=80
530 | num=9
531 | jitter=.3
532 | ignore_thresh = .7
533 | truth_thresh = 1
534 | scale_x_y = 1.2
535 | iou_thresh=0.213
536 | cls_normalizer=1.0
537 | iou_normalizer=0.07
538 | iou_loss=ciou
539 | nms_kind=greedynms
540 | beta_nms=0.6
541 | 
542 | [route]
543 | layers = -3
544 | 
545 | [convolutional]
546 | batch_normalize=1
547 | filters=128
548 | size=3
549 | stride=2
550 | pad=1
551 | activation=SiLU
552 | 
553 | [route]
554 | layers = -1,-14
555 | 
556 | #C3
557 | [convolutional]
558 | batch_normalize=1
559 | filters=128
560 | size=1
561 | stride=1
562 | pad=1
563 | activation=SiLU
564 | 
565 | [route]
566 | layers = -2
567 | 
568 | [convolutional]
569 | batch_normalize=1
570 | filters=128
571 | size=1
572 | stride=1
573 | pad=1
574 | activation=SiLU
575 | 
576 | [convolutional]
577 | batch_normalize=1
578 | filters=128
579 | size=1
580 | stride=1
581 | pad=1
582 | activation=SiLU
583 | 
584 | [convolutional]
585 | batch_normalize=1
586 | filters=128
587 | size=3
588 | stride=1
589 | pad=1
590 | activation=SiLU
591 | 
592 | [route]
593 | layers = -1,-5
594 | 
595 | [convolutional]
596 | batch_normalize=1
597 | filters=256
598 | size=1
599 | stride=1
600 | pad=1
601 | activation=SiLU
602 | 
603 | ######################
604 | [convolutional]
605 | size=1
606 | stride=1
607 | pad=1
608 | filters=255
609 | activation=linear
610 | 
611 | [yolo]
612 | mask = 3,4,5
613 | anchors = 10,13, 16,30, 33,23, 30,61, 62,45, 59,119, 116,90, 156,198, 373,326
614 | classes=80
615 | num=9
616 | jitter=.3
617 | ignore_thresh = .7
618 | truth_thresh = 1
619 | scale_x_y = 1.2
620 | iou_thresh=0.213
621 | cls_normalizer=1.0
622 | iou_normalizer=0.07
623 | iou_loss=ciou
624 | nms_kind=greedynms
625 | beta_nms=0.6
626 | 
627 | [route]
628 | layers = -3
629 | 
630 | [convolutional]
631 | batch_normalize=1
632 | filters=256
633 | size=3
634 | stride=2
635 | pad=1
636 | activation=SiLU
637 | 
638 | [route]
639 | layers = -1,-36
640 | 
641 | #C3
642 | [convolutional]
643 | batch_normalize=1
644 | filters=256
645 | size=1
646 | stride=1
647 | pad=1
648 | activation=SiLU
649 | 
650 | [route]
651 | layers = -2
652 | 
653 | [convolutional]
654 | batch_normalize=1
655 | filters=256
656 | size=1
657 | stride=1
658 | pad=1
659 | activation=SiLU
660 | 
661 | [convolutional]
662 | batch_normalize=1
663 | filters=256
664 | size=1
665 | stride=1
666 | pad=1
667 | activation=SiLU
668 | 
669 | [convolutional]
670 | batch_normalize=1
671 | filters=256
672 | size=3
673 | stride=1
674 | pad=1
675 | activation=SiLU
676 | 
677 | [route]
678 | layers = -1,-5
679 | 
680 | [convolutional]
681 | batch_normalize=1
682 | filters=512
683 | size=1
684 | stride=1
685 | pad=1
686 | activation=SiLU
687 | 
688 | ######################
689 | [convolutional]
690 | size=1
691 | stride=1
692 | pad=1
693 | filters=255
694 | activation=linear
695 | 
696 | [yolo]
697 | mask = 6,7,8
698 | anchors = 10,13, 16,30, 33,23, 30,61, 62,45, 59,119, 116,90, 156,198, 373,326
699 | classes=80
700 | num=9
701 | jitter=.3
702 | ignore_thresh = .7
703 | truth_thresh = 1
704 | scale_x_y = 1.2
705 | iou_thresh=0.213
706 | cls_normalizer=1.0
707 | iou_normalizer=0.07
708 | iou_loss=ciou
709 | nms_kind=greedynms
710 | beta_nms=0.6
711 | 
712 | 
713 | 


--------------------------------------------------------------------------------
/cfg/0514/yolov5s_v4_hand.cfg:
--------------------------------------------------------------------------------
  1 | [net]
  2 | # Testing
  3 | #batch=1
  4 | #subdivisions=1
  5 | # Training
  6 | batch=64
  7 | subdivisions=8
  8 | width=608
  9 | height=608
 10 | channels=3
 11 | momentum=0.949
 12 | decay=0.0005
 13 | angle=0
 14 | saturation = 1.5
 15 | exposure = 1.5
 16 | hue=.1
 17 | 
 18 | learning_rate=0.00261
 19 | burn_in=1000
 20 | max_batches = 500500
 21 | policy=steps
 22 | steps=400000,450000
 23 | scales=.1,.1
 24 | 
 25 | #cutmix=1
 26 | mosaic=1
 27 | 
 28 | #:104x104 54:52x52 85:26x26 104:13x13 for 416
 29 | [focus]
 30 | filters=12
 31 | 
 32 | [convolutional]
 33 | batch_normalize=1
 34 | filters=32
 35 | size=3
 36 | stride=1
 37 | pad=1
 38 | activation=SiLU
 39 | 
 40 | # Downsample
 41 | [convolutional]
 42 | batch_normalize=1
 43 | filters=64
 44 | size=3
 45 | stride=2
 46 | pad=1
 47 | activation=SiLU
 48 | 
 49 | #C3
 50 | [convolutional]
 51 | batch_normalize=1
 52 | filters=32
 53 | size=1
 54 | stride=1
 55 | pad=1
 56 | activation=SiLU
 57 | 
 58 | [route]
 59 | layers = -2
 60 | 
 61 | [convolutional]
 62 | batch_normalize=1
 63 | filters=32
 64 | size=1
 65 | stride=1
 66 | pad=1
 67 | activation=SiLU
 68 | 
 69 | [convolutional]
 70 | batch_normalize=1
 71 | filters=32
 72 | size=1
 73 | stride=1
 74 | pad=1
 75 | activation=SiLU
 76 | 
 77 | [convolutional]
 78 | batch_normalize=1
 79 | filters=32
 80 | size=3
 81 | stride=1
 82 | pad=1
 83 | activation=SiLU
 84 | 
 85 | [shortcut]
 86 | from=-3
 87 | activation=linear
 88 | 
 89 | [route]
 90 | layers = -1,-6
 91 | 
 92 | [convolutional]
 93 | batch_normalize=1
 94 | filters=64
 95 | size=1
 96 | stride=1
 97 | pad=1
 98 | activation=SiLU
 99 | 
100 | # Downsample
101 | [convolutional]
102 | batch_normalize=1
103 | filters=128
104 | size=3
105 | stride=2
106 | pad=1
107 | activation=SiLU
108 | 
109 | #C3
110 | [convolutional]
111 | batch_normalize=1
112 | filters=64
113 | size=1
114 | stride=1
115 | pad=1
116 | activation=SiLU
117 | 
118 | [route]
119 | layers = -2
120 | 
121 | [convolutional]
122 | batch_normalize=1
123 | filters=64
124 | size=1
125 | stride=1
126 | pad=1
127 | activation=SiLU
128 | 
129 | [convolutional]
130 | batch_normalize=1
131 | filters=64
132 | size=1
133 | stride=1
134 | pad=1
135 | activation=SiLU
136 | 
137 | [convolutional]
138 | batch_normalize=1
139 | filters=64
140 | size=3
141 | stride=1
142 | pad=1
143 | activation=SiLU
144 | 
145 | [shortcut]
146 | from=-3
147 | activation=linear
148 | 
149 | [convolutional]
150 | batch_normalize=1
151 | filters=64
152 | size=1
153 | stride=1
154 | pad=1
155 | activation=SiLU
156 | 
157 | [convolutional]
158 | batch_normalize=1
159 | filters=64
160 | size=3
161 | stride=1
162 | pad=1
163 | activation=SiLU
164 | 
165 | [shortcut]
166 | from=-3
167 | activation=linear
168 | 
169 | [convolutional]
170 | batch_normalize=1
171 | filters=64
172 | size=1
173 | stride=1
174 | pad=1
175 | activation=SiLU
176 | 
177 | [convolutional]
178 | batch_normalize=1
179 | filters=64
180 | size=3
181 | stride=1
182 | pad=1
183 | activation=SiLU
184 | 
185 | [shortcut]
186 | from=-3
187 | activation=linear
188 | 
189 | [route]
190 | layers = -1,-12
191 | 
192 | [convolutional]
193 | batch_normalize=1
194 | filters=128
195 | size=1
196 | stride=1
197 | pad=1
198 | activation=SiLU
199 | 
200 | # Downsample
201 | [convolutional]
202 | batch_normalize=1
203 | filters=256
204 | size=3
205 | stride=2
206 | pad=1
207 | activation=SiLU
208 | 
209 | #C3
210 | [convolutional]
211 | batch_normalize=1
212 | filters=128
213 | size=1
214 | stride=1
215 | pad=1
216 | activation=SiLU
217 | 
218 | [route]
219 | layers = -2
220 | 
221 | [convolutional]
222 | batch_normalize=1
223 | filters=128
224 | size=1
225 | stride=1
226 | pad=1
227 | activation=SiLU
228 | 
229 | [convolutional]
230 | batch_normalize=1
231 | filters=128
232 | size=1
233 | stride=1
234 | pad=1
235 | activation=SiLU
236 | 
237 | [convolutional]
238 | batch_normalize=1
239 | filters=128
240 | size=3
241 | stride=1
242 | pad=1
243 | activation=SiLU
244 | 
245 | [shortcut]
246 | from=-3
247 | activation=linear
248 | 
249 | [convolutional]
250 | batch_normalize=1
251 | filters=128
252 | size=1
253 | stride=1
254 | pad=1
255 | activation=SiLU
256 | 
257 | [convolutional]
258 | batch_normalize=1
259 | filters=128
260 | size=3
261 | stride=1
262 | pad=1
263 | activation=SiLU
264 | 
265 | [shortcut]
266 | from=-3
267 | activation=linear
268 | 
269 | [convolutional]
270 | batch_normalize=1
271 | filters=128
272 | size=1
273 | stride=1
274 | pad=1
275 | activation=SiLU
276 | 
277 | [convolutional]
278 | batch_normalize=1
279 | filters=128
280 | size=3
281 | stride=1
282 | pad=1
283 | activation=SiLU
284 | 
285 | [shortcut]
286 | from=-3
287 | activation=linear
288 | 
289 | [route]
290 | layers = -1,-12
291 | 
292 | [convolutional]
293 | batch_normalize=1
294 | filters=256
295 | size=1
296 | stride=1
297 | pad=1
298 | activation=SiLU
299 | 
300 | # Downsample
301 | [convolutional]
302 | batch_normalize=1
303 | filters=512
304 | size=3
305 | stride=2
306 | pad=1
307 | activation=SiLU
308 | 
309 | [convolutional]
310 | batch_normalize=1
311 | filters=256
312 | size=1
313 | stride=1
314 | pad=1
315 | activation=SiLU
316 | 
317 | ### SPP ###
318 | [maxpool]
319 | stride=1
320 | size=5
321 | 
322 | [route]
323 | layers=-2
324 | 
325 | [maxpool]
326 | stride=1
327 | size=9
328 | 
329 | [route]
330 | layers=-4
331 | 
332 | [maxpool]
333 | stride=1
334 | size=13
335 | 
336 | [route]
337 | ###layers=-1,-3,-5,-6
338 | layers=-6,-5,-3,-1
339 | ### End SPP ###
340 | 
341 | [convolutional]
342 | batch_normalize=1
343 | filters=512
344 | size=1
345 | stride=1
346 | pad=1
347 | activation=SiLU
348 | 
349 | #C3
350 | [convolutional]
351 | batch_normalize=1
352 | filters=256
353 | size=1
354 | stride=1
355 | pad=1
356 | activation=SiLU
357 | 
358 | [route]
359 | layers = -2
360 | 
361 | [convolutional]
362 | batch_normalize=1
363 | filters=256
364 | size=1
365 | stride=1
366 | pad=1
367 | activation=SiLU
368 | 
369 | [convolutional]
370 | batch_normalize=1
371 | filters=256
372 | size=1
373 | stride=1
374 | pad=1
375 | activation=SiLU
376 | 
377 | [convolutional]
378 | batch_normalize=1
379 | filters=256
380 | size=3
381 | stride=1
382 | pad=1
383 | activation=SiLU
384 | 
385 | [route]
386 | layers = -1,-5
387 | 
388 | [convolutional]
389 | batch_normalize=1
390 | filters=512
391 | size=1
392 | stride=1
393 | pad=1
394 | activation=SiLU
395 | 
396 | [convolutional]
397 | batch_normalize=1
398 | filters=256
399 | size=1
400 | stride=1
401 | pad=1
402 | activation=SiLU
403 | 
404 | [upsample]
405 | stride=2
406 | 
407 | [route]
408 | layers = -1,-19
409 | 
410 | #C3
411 | [convolutional]
412 | batch_normalize=1
413 | filters=128
414 | size=1
415 | stride=1
416 | pad=1
417 | activation=SiLU
418 | 
419 | [route]
420 | layers = -2
421 | 
422 | [convolutional]
423 | batch_normalize=1
424 | filters=128
425 | size=1
426 | stride=1
427 | pad=1
428 | activation=SiLU
429 | 
430 | [convolutional]
431 | batch_normalize=1
432 | filters=128
433 | size=1
434 | stride=1
435 | pad=1
436 | activation=SiLU
437 | 
438 | [convolutional]
439 | batch_normalize=1
440 | filters=128
441 | size=3
442 | stride=1
443 | pad=1
444 | activation=SiLU
445 | 
446 | [route]
447 | layers = -1,-5
448 | 
449 | [convolutional]
450 | batch_normalize=1
451 | filters=256
452 | size=1
453 | stride=1
454 | pad=1
455 | activation=SiLU
456 | 
457 | [convolutional]
458 | batch_normalize=1
459 | filters=128
460 | size=1
461 | stride=1
462 | pad=1
463 | activation=SiLU
464 | 
465 | [upsample]
466 | stride=2
467 | 
468 | [route]
469 | layers = -1,-44
470 | 
471 | #C3
472 | [convolutional]
473 | batch_normalize=1
474 | filters=64
475 | size=1
476 | stride=1
477 | pad=1
478 | activation=SiLU
479 | 
480 | [route]
481 | layers = -2
482 | 
483 | [convolutional]
484 | batch_normalize=1
485 | filters=64
486 | size=1
487 | stride=1
488 | pad=1
489 | activation=SiLU
490 | 
491 | [convolutional]
492 | batch_normalize=1
493 | filters=64
494 | size=1
495 | stride=1
496 | pad=1
497 | activation=SiLU
498 | 
499 | [convolutional]
500 | batch_normalize=1
501 | filters=64
502 | size=3
503 | stride=1
504 | pad=1
505 | activation=SiLU
506 | 
507 | [route]
508 | layers = -1,-5
509 | 
510 | [convolutional]
511 | batch_normalize=1
512 | filters=128
513 | size=1
514 | stride=1
515 | pad=1
516 | activation=SiLU
517 | 
518 | ######################
519 | [convolutional]
520 | size=1
521 | stride=1
522 | pad=1
523 | filters=18
524 | activation=linear
525 | 
526 | [yolo]
527 | mask = 0,1,2
528 | anchors = 10,13, 16,30, 33,23, 30,61, 62,45, 59,119, 116,90, 156,198, 373,326
529 | classes=1
530 | num=9
531 | jitter=.3
532 | ignore_thresh = .7
533 | truth_thresh = 1
534 | scale_x_y = 1.2
535 | iou_thresh=0.213
536 | cls_normalizer=1.0
537 | iou_normalizer=0.07
538 | iou_loss=ciou
539 | nms_kind=greedynms
540 | beta_nms=0.6
541 | 
542 | [route]
543 | layers = -3
544 | 
545 | [convolutional]
546 | batch_normalize=1
547 | filters=128
548 | size=3
549 | stride=2
550 | pad=1
551 | activation=SiLU
552 | 
553 | [route]
554 | layers = -1,-14
555 | 
556 | #C3
557 | [convolutional]
558 | batch_normalize=1
559 | filters=128
560 | size=1
561 | stride=1
562 | pad=1
563 | activation=SiLU
564 | 
565 | [route]
566 | layers = -2
567 | 
568 | [convolutional]
569 | batch_normalize=1
570 | filters=128
571 | size=1
572 | stride=1
573 | pad=1
574 | activation=SiLU
575 | 
576 | [convolutional]
577 | batch_normalize=1
578 | filters=128
579 | size=1
580 | stride=1
581 | pad=1
582 | activation=SiLU
583 | 
584 | [convolutional]
585 | batch_normalize=1
586 | filters=128
587 | size=3
588 | stride=1
589 | pad=1
590 | activation=SiLU
591 | 
592 | [route]
593 | layers = -1,-5
594 | 
595 | [convolutional]
596 | batch_normalize=1
597 | filters=256
598 | size=1
599 | stride=1
600 | pad=1
601 | activation=SiLU
602 | 
603 | ######################
604 | [convolutional]
605 | size=1
606 | stride=1
607 | pad=1
608 | filters=18
609 | activation=linear
610 | 
611 | [yolo]
612 | mask = 3,4,5
613 | anchors = 10,13, 16,30, 33,23, 30,61, 62,45, 59,119, 116,90, 156,198, 373,326
614 | classes=1
615 | num=9
616 | jitter=.3
617 | ignore_thresh = .7
618 | truth_thresh = 1
619 | scale_x_y = 1.2
620 | iou_thresh=0.213
621 | cls_normalizer=1.0
622 | iou_normalizer=0.07
623 | iou_loss=ciou
624 | nms_kind=greedynms
625 | beta_nms=0.6
626 | 
627 | [route]
628 | layers = -3
629 | 
630 | [convolutional]
631 | batch_normalize=1
632 | filters=256
633 | size=3
634 | stride=2
635 | pad=1
636 | activation=SiLU
637 | 
638 | [route]
639 | layers = -1,-36
640 | 
641 | #C3
642 | [convolutional]
643 | batch_normalize=1
644 | filters=256
645 | size=1
646 | stride=1
647 | pad=1
648 | activation=SiLU
649 | 
650 | [route]
651 | layers = -2
652 | 
653 | [convolutional]
654 | batch_normalize=1
655 | filters=256
656 | size=1
657 | stride=1
658 | pad=1
659 | activation=SiLU
660 | 
661 | [convolutional]
662 | batch_normalize=1
663 | filters=256
664 | size=1
665 | stride=1
666 | pad=1
667 | activation=SiLU
668 | 
669 | [convolutional]
670 | batch_normalize=1
671 | filters=256
672 | size=3
673 | stride=1
674 | pad=1
675 | activation=SiLU
676 | 
677 | [route]
678 | layers = -1,-5
679 | 
680 | [convolutional]
681 | batch_normalize=1
682 | filters=512
683 | size=1
684 | stride=1
685 | pad=1
686 | activation=SiLU
687 | 
688 | ######################
689 | [convolutional]
690 | size=1
691 | stride=1
692 | pad=1
693 | filters=18
694 | activation=linear
695 | 
696 | [yolo]
697 | mask = 6,7,8
698 | anchors = 10,13, 16,30, 33,23, 30,61, 62,45, 59,119, 116,90, 156,198, 373,326
699 | classes=1
700 | num=9
701 | jitter=.3
702 | ignore_thresh = .7
703 | truth_thresh = 1
704 | scale_x_y = 1.2
705 | iou_thresh=0.213
706 | cls_normalizer=1.0
707 | iou_normalizer=0.07
708 | iou_loss=ciou
709 | nms_kind=greedynms
710 | beta_nms=0.6
711 | 
712 | 
713 | 


--------------------------------------------------------------------------------
/cfg/last_prune/prune_0.8_keep_0.01_8x_yolov5l_v4.cfg:
--------------------------------------------------------------------------------
   1 | [net]
   2 | batch=64
   3 | subdivisions=8
   4 | width=416
   5 | height=416
   6 | channels=3
   7 | momentum=0.949
   8 | decay=0.0005
   9 | angle=0
  10 | saturation=1.5
  11 | exposure=1.5
  12 | hue=.1
  13 | learning_rate=0.00261
  14 | burn_in=1000
  15 | max_batches=500500
  16 | policy=steps
  17 | steps=400000,450000
  18 | scales=.1,.1
  19 | mosaic=1
  20 | 
  21 | [focus]
  22 | filters=12
  23 | 
  24 | [convolutional]
  25 | batch_normalize=1
  26 | filters=40
  27 | size=3
  28 | stride=1
  29 | pad=1
  30 | activation=leaky
  31 | 
  32 | [convolutional]
  33 | batch_normalize=1
  34 | filters=112
  35 | size=3
  36 | stride=2
  37 | pad=1
  38 | activation=leaky
  39 | 
  40 | [convolutional]
  41 | batch_normalize=1
  42 | filters=64
  43 | size=1
  44 | stride=1
  45 | pad=1
  46 | activation=leaky
  47 | 
  48 | [route]
  49 | layers=-2
  50 | 
  51 | [convolutional]
  52 | batch_normalize=1
  53 | filters=64
  54 | size=1
  55 | stride=1
  56 | pad=1
  57 | activation=leaky
  58 | 
  59 | [convolutional]
  60 | batch_normalize=1
  61 | filters=64
  62 | size=1
  63 | stride=1
  64 | pad=1
  65 | activation=leaky
  66 | 
  67 | [convolutional]
  68 | batch_normalize=1
  69 | filters=64
  70 | size=3
  71 | stride=1
  72 | pad=1
  73 | activation=leaky
  74 | 
  75 | [shortcut]
  76 | from=-3
  77 | activation=linear
  78 | 
  79 | [convolutional]
  80 | batch_normalize=1
  81 | filters=64
  82 | size=1
  83 | stride=1
  84 | pad=1
  85 | activation=leaky
  86 | 
  87 | [convolutional]
  88 | batch_normalize=1
  89 | filters=64
  90 | size=3
  91 | stride=1
  92 | pad=1
  93 | activation=leaky
  94 | 
  95 | [shortcut]
  96 | from=-3
  97 | activation=linear
  98 | 
  99 | [convolutional]
 100 | batch_normalize=1
 101 | filters=64
 102 | size=1
 103 | stride=1
 104 | pad=1
 105 | activation=leaky
 106 | 
 107 | [convolutional]
 108 | batch_normalize=1
 109 | filters=64
 110 | size=3
 111 | stride=1
 112 | pad=1
 113 | activation=leaky
 114 | 
 115 | [shortcut]
 116 | from=-3
 117 | activation=linear
 118 | 
 119 | [route]
 120 | layers=-1,-12
 121 | 
 122 | [convolutional]
 123 | batch_normalize=1
 124 | filters=128
 125 | size=1
 126 | stride=1
 127 | pad=1
 128 | activation=leaky
 129 | 
 130 | [convolutional]
 131 | batch_normalize=1
 132 | filters=248
 133 | size=3
 134 | stride=2
 135 | pad=1
 136 | activation=leaky
 137 | 
 138 | [convolutional]
 139 | batch_normalize=1
 140 | filters=128
 141 | size=1
 142 | stride=1
 143 | pad=1
 144 | activation=leaky
 145 | 
 146 | [route]
 147 | layers=-2
 148 | 
 149 | [convolutional]
 150 | batch_normalize=1
 151 | filters=128
 152 | size=1
 153 | stride=1
 154 | pad=1
 155 | activation=leaky
 156 | 
 157 | [convolutional]
 158 | batch_normalize=1
 159 | filters=56
 160 | size=1
 161 | stride=1
 162 | pad=1
 163 | activation=leaky
 164 | 
 165 | [convolutional]
 166 | batch_normalize=1
 167 | filters=128
 168 | size=3
 169 | stride=1
 170 | pad=1
 171 | activation=leaky
 172 | 
 173 | [shortcut]
 174 | from=-3
 175 | activation=linear
 176 | 
 177 | [convolutional]
 178 | batch_normalize=1
 179 | filters=56
 180 | size=1
 181 | stride=1
 182 | pad=1
 183 | activation=leaky
 184 | 
 185 | [convolutional]
 186 | batch_normalize=1
 187 | filters=128
 188 | size=3
 189 | stride=1
 190 | pad=1
 191 | activation=leaky
 192 | 
 193 | [shortcut]
 194 | from=-3
 195 | activation=linear
 196 | 
 197 | [convolutional]
 198 | batch_normalize=1
 199 | filters=24
 200 | size=1
 201 | stride=1
 202 | pad=1
 203 | activation=leaky
 204 | 
 205 | [convolutional]
 206 | batch_normalize=1
 207 | filters=128
 208 | size=3
 209 | stride=1
 210 | pad=1
 211 | activation=leaky
 212 | 
 213 | [shortcut]
 214 | from=-3
 215 | activation=linear
 216 | 
 217 | [convolutional]
 218 | batch_normalize=1
 219 | filters=24
 220 | size=1
 221 | stride=1
 222 | pad=1
 223 | activation=leaky
 224 | 
 225 | [convolutional]
 226 | batch_normalize=1
 227 | filters=128
 228 | size=3
 229 | stride=1
 230 | pad=1
 231 | activation=leaky
 232 | 
 233 | [shortcut]
 234 | from=-3
 235 | activation=linear
 236 | 
 237 | [convolutional]
 238 | batch_normalize=1
 239 | filters=40
 240 | size=1
 241 | stride=1
 242 | pad=1
 243 | activation=leaky
 244 | 
 245 | [convolutional]
 246 | batch_normalize=1
 247 | filters=128
 248 | size=3
 249 | stride=1
 250 | pad=1
 251 | activation=leaky
 252 | 
 253 | [shortcut]
 254 | from=-3
 255 | activation=linear
 256 | 
 257 | [convolutional]
 258 | batch_normalize=1
 259 | filters=32
 260 | size=1
 261 | stride=1
 262 | pad=1
 263 | activation=leaky
 264 | 
 265 | [convolutional]
 266 | batch_normalize=1
 267 | filters=128
 268 | size=3
 269 | stride=1
 270 | pad=1
 271 | activation=leaky
 272 | 
 273 | [shortcut]
 274 | from=-3
 275 | activation=linear
 276 | 
 277 | [convolutional]
 278 | batch_normalize=1
 279 | filters=64
 280 | size=1
 281 | stride=1
 282 | pad=1
 283 | activation=leaky
 284 | 
 285 | [convolutional]
 286 | batch_normalize=1
 287 | filters=128
 288 | size=3
 289 | stride=1
 290 | pad=1
 291 | activation=leaky
 292 | 
 293 | [shortcut]
 294 | from=-3
 295 | activation=linear
 296 | 
 297 | [convolutional]
 298 | batch_normalize=1
 299 | filters=64
 300 | size=1
 301 | stride=1
 302 | pad=1
 303 | activation=leaky
 304 | 
 305 | [convolutional]
 306 | batch_normalize=1
 307 | filters=128
 308 | size=3
 309 | stride=1
 310 | pad=1
 311 | activation=leaky
 312 | 
 313 | [shortcut]
 314 | from=-3
 315 | activation=linear
 316 | 
 317 | [convolutional]
 318 | batch_normalize=1
 319 | filters=56
 320 | size=1
 321 | stride=1
 322 | pad=1
 323 | activation=leaky
 324 | 
 325 | [convolutional]
 326 | batch_normalize=1
 327 | filters=128
 328 | size=3
 329 | stride=1
 330 | pad=1
 331 | activation=leaky
 332 | 
 333 | [shortcut]
 334 | from=-3
 335 | activation=linear
 336 | 
 337 | [route]
 338 | layers=-1,-30
 339 | 
 340 | [convolutional]
 341 | batch_normalize=1
 342 | filters=240
 343 | size=1
 344 | stride=1
 345 | pad=1
 346 | activation=leaky
 347 | 
 348 | [convolutional]
 349 | batch_normalize=1
 350 | filters=240
 351 | size=3
 352 | stride=2
 353 | pad=1
 354 | activation=leaky
 355 | 
 356 | [convolutional]
 357 | batch_normalize=1
 358 | filters=96
 359 | size=1
 360 | stride=1
 361 | pad=1
 362 | activation=leaky
 363 | 
 364 | [route]
 365 | layers=-2
 366 | 
 367 | [convolutional]
 368 | batch_normalize=1
 369 | filters=245
 370 | size=1
 371 | stride=1
 372 | pad=1
 373 | activation=leaky
 374 | 
 375 | [convolutional]
 376 | batch_normalize=1
 377 | filters=56
 378 | size=1
 379 | stride=1
 380 | pad=1
 381 | activation=leaky
 382 | 
 383 | [convolutional]
 384 | batch_normalize=1
 385 | filters=245
 386 | size=3
 387 | stride=1
 388 | pad=1
 389 | activation=leaky
 390 | 
 391 | [shortcut]
 392 | from=-3
 393 | activation=linear
 394 | 
 395 | [convolutional]
 396 | batch_normalize=1
 397 | filters=8
 398 | size=1
 399 | stride=1
 400 | pad=1
 401 | activation=leaky
 402 | 
 403 | [convolutional]
 404 | batch_normalize=1
 405 | filters=245
 406 | size=3
 407 | stride=1
 408 | pad=1
 409 | activation=leaky
 410 | 
 411 | [shortcut]
 412 | from=-3
 413 | activation=linear
 414 | 
 415 | [convolutional]
 416 | batch_normalize=1
 417 | filters=16
 418 | size=1
 419 | stride=1
 420 | pad=1
 421 | activation=leaky
 422 | 
 423 | [convolutional]
 424 | batch_normalize=1
 425 | filters=245
 426 | size=3
 427 | stride=1
 428 | pad=1
 429 | activation=leaky
 430 | 
 431 | [shortcut]
 432 | from=-3
 433 | activation=linear
 434 | 
 435 | [convolutional]
 436 | batch_normalize=1
 437 | filters=24
 438 | size=1
 439 | stride=1
 440 | pad=1
 441 | activation=leaky
 442 | 
 443 | [convolutional]
 444 | batch_normalize=1
 445 | filters=245
 446 | size=3
 447 | stride=1
 448 | pad=1
 449 | activation=leaky
 450 | 
 451 | [shortcut]
 452 | from=-3
 453 | activation=linear
 454 | 
 455 | [convolutional]
 456 | batch_normalize=1
 457 | filters=16
 458 | size=1
 459 | stride=1
 460 | pad=1
 461 | activation=leaky
 462 | 
 463 | [convolutional]
 464 | batch_normalize=1
 465 | filters=245
 466 | size=3
 467 | stride=1
 468 | pad=1
 469 | activation=leaky
 470 | 
 471 | [shortcut]
 472 | from=-3
 473 | activation=linear
 474 | 
 475 | [convolutional]
 476 | batch_normalize=1
 477 | filters=24
 478 | size=1
 479 | stride=1
 480 | pad=1
 481 | activation=leaky
 482 | 
 483 | [convolutional]
 484 | batch_normalize=1
 485 | filters=245
 486 | size=3
 487 | stride=1
 488 | pad=1
 489 | activation=leaky
 490 | 
 491 | [shortcut]
 492 | from=-3
 493 | activation=linear
 494 | 
 495 | [convolutional]
 496 | batch_normalize=1
 497 | filters=16
 498 | size=1
 499 | stride=1
 500 | pad=1
 501 | activation=leaky
 502 | 
 503 | [convolutional]
 504 | batch_normalize=1
 505 | filters=245
 506 | size=3
 507 | stride=1
 508 | pad=1
 509 | activation=leaky
 510 | 
 511 | [shortcut]
 512 | from=-3
 513 | activation=linear
 514 | 
 515 | [convolutional]
 516 | batch_normalize=1
 517 | filters=32
 518 | size=1
 519 | stride=1
 520 | pad=1
 521 | activation=leaky
 522 | 
 523 | [convolutional]
 524 | batch_normalize=1
 525 | filters=245
 526 | size=3
 527 | stride=1
 528 | pad=1
 529 | activation=leaky
 530 | 
 531 | [shortcut]
 532 | from=-3
 533 | activation=linear
 534 | 
 535 | [convolutional]
 536 | batch_normalize=1
 537 | filters=24
 538 | size=1
 539 | stride=1
 540 | pad=1
 541 | activation=leaky
 542 | 
 543 | [convolutional]
 544 | batch_normalize=1
 545 | filters=245
 546 | size=3
 547 | stride=1
 548 | pad=1
 549 | activation=leaky
 550 | 
 551 | [shortcut]
 552 | from=-3
 553 | activation=linear
 554 | 
 555 | [route]
 556 | layers=-1,-30
 557 | 
 558 | [convolutional]
 559 | batch_normalize=1
 560 | filters=144
 561 | size=1
 562 | stride=1
 563 | pad=1
 564 | activation=leaky
 565 | 
 566 | [convolutional]
 567 | batch_normalize=1
 568 | filters=104
 569 | size=3
 570 | stride=2
 571 | pad=1
 572 | activation=leaky
 573 | 
 574 | [convolutional]
 575 | batch_normalize=1
 576 | filters=512
 577 | size=1
 578 | stride=1
 579 | pad=1
 580 | activation=leaky
 581 | 
 582 | [maxpool]
 583 | stride=1
 584 | size=5
 585 | 
 586 | [route]
 587 | layers=-2
 588 | 
 589 | [maxpool]
 590 | stride=1
 591 | size=9
 592 | 
 593 | [route]
 594 | layers=-4
 595 | 
 596 | [maxpool]
 597 | stride=1
 598 | size=13
 599 | 
 600 | [route]
 601 | layers=-6,-5,-3,-1
 602 | 
 603 | [convolutional]
 604 | batch_normalize=1
 605 | filters=56
 606 | size=1
 607 | stride=1
 608 | pad=1
 609 | activation=leaky
 610 | 
 611 | [convolutional]
 612 | batch_normalize=1
 613 | filters=40
 614 | size=1
 615 | stride=1
 616 | pad=1
 617 | activation=leaky
 618 | 
 619 | [route]
 620 | layers=-2
 621 | 
 622 | [convolutional]
 623 | batch_normalize=1
 624 | filters=8
 625 | size=1
 626 | stride=1
 627 | pad=1
 628 | activation=leaky
 629 | 
 630 | [convolutional]
 631 | batch_normalize=1
 632 | filters=8
 633 | size=1
 634 | stride=1
 635 | pad=1
 636 | activation=leaky
 637 | 
 638 | [convolutional]
 639 | batch_normalize=1
 640 | filters=8
 641 | size=3
 642 | stride=1
 643 | pad=1
 644 | activation=leaky
 645 | 
 646 | [convolutional]
 647 | batch_normalize=1
 648 | filters=8
 649 | size=1
 650 | stride=1
 651 | pad=1
 652 | activation=leaky
 653 | 
 654 | [convolutional]
 655 | batch_normalize=1
 656 | filters=16
 657 | size=3
 658 | stride=1
 659 | pad=1
 660 | activation=leaky
 661 | 
 662 | [convolutional]
 663 | batch_normalize=1
 664 | filters=8
 665 | size=1
 666 | stride=1
 667 | pad=1
 668 | activation=leaky
 669 | 
 670 | [convolutional]
 671 | batch_normalize=1
 672 | filters=8
 673 | size=3
 674 | stride=1
 675 | pad=1
 676 | activation=leaky
 677 | 
 678 | [route]
 679 | layers=-1,-9
 680 | 
 681 | [convolutional]
 682 | batch_normalize=1
 683 | filters=40
 684 | size=1
 685 | stride=1
 686 | pad=1
 687 | activation=leaky
 688 | 
 689 | [convolutional]
 690 | batch_normalize=1
 691 | filters=512
 692 | size=1
 693 | stride=1
 694 | pad=1
 695 | activation=leaky
 696 | 
 697 | [upsample]
 698 | stride=2
 699 | 
 700 | [route]
 701 | layers=-1,-23
 702 | 
 703 | [convolutional]
 704 | batch_normalize=1
 705 | filters=48
 706 | size=1
 707 | stride=1
 708 | pad=1
 709 | activation=leaky
 710 | 
 711 | [route]
 712 | layers=-2
 713 | 
 714 | [convolutional]
 715 | batch_normalize=1
 716 | filters=80
 717 | size=1
 718 | stride=1
 719 | pad=1
 720 | activation=leaky
 721 | 
 722 | [convolutional]
 723 | batch_normalize=1
 724 | filters=104
 725 | size=1
 726 | stride=1
 727 | pad=1
 728 | activation=leaky
 729 | 
 730 | [convolutional]
 731 | batch_normalize=1
 732 | filters=128
 733 | size=3
 734 | stride=1
 735 | pad=1
 736 | activation=leaky
 737 | 
 738 | [convolutional]
 739 | batch_normalize=1
 740 | filters=112
 741 | size=1
 742 | stride=1
 743 | pad=1
 744 | activation=leaky
 745 | 
 746 | [convolutional]
 747 | batch_normalize=1
 748 | filters=112
 749 | size=3
 750 | stride=1
 751 | pad=1
 752 | activation=leaky
 753 | 
 754 | [convolutional]
 755 | batch_normalize=1
 756 | filters=96
 757 | size=1
 758 | stride=1
 759 | pad=1
 760 | activation=leaky
 761 | 
 762 | [convolutional]
 763 | batch_normalize=1
 764 | filters=104
 765 | size=3
 766 | stride=1
 767 | pad=1
 768 | activation=leaky
 769 | 
 770 | [route]
 771 | layers=-1,-9
 772 | 
 773 | [convolutional]
 774 | batch_normalize=1
 775 | filters=88
 776 | size=1
 777 | stride=1
 778 | pad=1
 779 | activation=leaky
 780 | 
 781 | [convolutional]
 782 | batch_normalize=1
 783 | filters=256
 784 | size=1
 785 | stride=1
 786 | pad=1
 787 | activation=leaky
 788 | 
 789 | [upsample]
 790 | stride=2
 791 | 
 792 | [route]
 793 | layers=-1,-70
 794 | 
 795 | [convolutional]
 796 | batch_normalize=1
 797 | filters=32
 798 | size=1
 799 | stride=1
 800 | pad=1
 801 | activation=leaky
 802 | 
 803 | [route]
 804 | layers=-2
 805 | 
 806 | [convolutional]
 807 | batch_normalize=1
 808 | filters=72
 809 | size=1
 810 | stride=1
 811 | pad=1
 812 | activation=leaky
 813 | 
 814 | [convolutional]
 815 | batch_normalize=1
 816 | filters=72
 817 | size=1
 818 | stride=1
 819 | pad=1
 820 | activation=leaky
 821 | 
 822 | [convolutional]
 823 | batch_normalize=1
 824 | filters=96
 825 | size=3
 826 | stride=1
 827 | pad=1
 828 | activation=leaky
 829 | 
 830 | [convolutional]
 831 | batch_normalize=1
 832 | filters=88
 833 | size=1
 834 | stride=1
 835 | pad=1
 836 | activation=leaky
 837 | 
 838 | [convolutional]
 839 | batch_normalize=1
 840 | filters=88
 841 | size=3
 842 | stride=1
 843 | pad=1
 844 | activation=leaky
 845 | 
 846 | [convolutional]
 847 | batch_normalize=1
 848 | filters=80
 849 | size=1
 850 | stride=1
 851 | pad=1
 852 | activation=leaky
 853 | 
 854 | [convolutional]
 855 | batch_normalize=1
 856 | filters=96
 857 | size=3
 858 | stride=1
 859 | pad=1
 860 | activation=leaky
 861 | 
 862 | [route]
 863 | layers=-1,-9
 864 | 
 865 | [convolutional]
 866 | batch_normalize=1
 867 | filters=96
 868 | size=1
 869 | stride=1
 870 | pad=1
 871 | activation=leaky
 872 | 
 873 | [convolutional]
 874 | size=1
 875 | stride=1
 876 | pad=1
 877 | filters=24
 878 | activation=linear
 879 | 
 880 | [yolo]
 881 | mask=0,1,2
 882 | anchors=40,39, 51,50, 61,59, 75,69, 62,92, 88,98, 115,77, 93,129, 128,115
 883 | classes=3
 884 | num=9
 885 | jitter=.3
 886 | ignore_thresh=.7
 887 | truth_thresh=1
 888 | scale_x_y=1.2
 889 | iou_thresh=0.213
 890 | cls_normalizer=1.0
 891 | iou_normalizer=0.07
 892 | iou_loss=ciou
 893 | nms_kind=greedynms
 894 | beta_nms=0.6
 895 | 
 896 | [route]
 897 | layers=-3
 898 | 
 899 | [convolutional]
 900 | batch_normalize=1
 901 | filters=72
 902 | size=3
 903 | stride=2
 904 | pad=1
 905 | activation=leaky
 906 | 
 907 | [route]
 908 | layers=-1,-18
 909 | 
 910 | [convolutional]
 911 | batch_normalize=1
 912 | filters=64
 913 | size=1
 914 | stride=1
 915 | pad=1
 916 | activation=leaky
 917 | 
 918 | [route]
 919 | layers=-2
 920 | 
 921 | [convolutional]
 922 | batch_normalize=1
 923 | filters=56
 924 | size=1
 925 | stride=1
 926 | pad=1
 927 | activation=leaky
 928 | 
 929 | [convolutional]
 930 | batch_normalize=1
 931 | filters=48
 932 | size=1
 933 | stride=1
 934 | pad=1
 935 | activation=leaky
 936 | 
 937 | [convolutional]
 938 | batch_normalize=1
 939 | filters=64
 940 | size=3
 941 | stride=1
 942 | pad=1
 943 | activation=leaky
 944 | 
 945 | [convolutional]
 946 | batch_normalize=1
 947 | filters=48
 948 | size=1
 949 | stride=1
 950 | pad=1
 951 | activation=leaky
 952 | 
 953 | [convolutional]
 954 | batch_normalize=1
 955 | filters=64
 956 | size=3
 957 | stride=1
 958 | pad=1
 959 | activation=leaky
 960 | 
 961 | [convolutional]
 962 | batch_normalize=1
 963 | filters=64
 964 | size=1
 965 | stride=1
 966 | pad=1
 967 | activation=leaky
 968 | 
 969 | [convolutional]
 970 | batch_normalize=1
 971 | filters=80
 972 | size=3
 973 | stride=1
 974 | pad=1
 975 | activation=leaky
 976 | 
 977 | [route]
 978 | layers=-1,-9
 979 | 
 980 | [convolutional]
 981 | batch_normalize=1
 982 | filters=104
 983 | size=1
 984 | stride=1
 985 | pad=1
 986 | activation=leaky
 987 | 
 988 | [convolutional]
 989 | size=1
 990 | stride=1
 991 | pad=1
 992 | filters=24
 993 | activation=linear
 994 | 
 995 | [yolo]
 996 | mask=3,4,5
 997 | anchors=40,39, 51,50, 61,59, 75,69, 62,92, 88,98, 115,77, 93,129, 128,115
 998 | classes=3
 999 | num=9
1000 | jitter=.3
1001 | ignore_thresh=.7
1002 | truth_thresh=1
1003 | scale_x_y=1.2
1004 | iou_thresh=0.213
1005 | cls_normalizer=1.0
1006 | iou_normalizer=0.07
1007 | iou_loss=ciou
1008 | nms_kind=greedynms
1009 | beta_nms=0.6
1010 | 
1011 | [route]
1012 | layers=-3
1013 | 
1014 | [convolutional]
1015 | batch_normalize=1
1016 | filters=136
1017 | size=3
1018 | stride=2
1019 | pad=1
1020 | activation=leaky
1021 | 
1022 | [route]
1023 | layers=-1,-48
1024 | 
1025 | [convolutional]
1026 | batch_normalize=1
1027 | filters=64
1028 | size=1
1029 | stride=1
1030 | pad=1
1031 | activation=leaky
1032 | 
1033 | [route]
1034 | layers=-2
1035 | 
1036 | [convolutional]
1037 | batch_normalize=1
1038 | filters=8
1039 | size=1
1040 | stride=1
1041 | pad=1
1042 | activation=leaky
1043 | 
1044 | [convolutional]
1045 | batch_normalize=1
1046 | filters=8
1047 | size=1
1048 | stride=1
1049 | pad=1
1050 | activation=leaky
1051 | 
1052 | [convolutional]
1053 | batch_normalize=1
1054 | filters=8
1055 | size=3
1056 | stride=1
1057 | pad=1
1058 | activation=leaky
1059 | 
1060 | [convolutional]
1061 | batch_normalize=1
1062 | filters=8
1063 | size=1
1064 | stride=1
1065 | pad=1
1066 | activation=leaky
1067 | 
1068 | [convolutional]
1069 | batch_normalize=1
1070 | filters=8
1071 | size=3
1072 | stride=1
1073 | pad=1
1074 | activation=leaky
1075 | 
1076 | [convolutional]
1077 | batch_normalize=1
1078 | filters=8
1079 | size=1
1080 | stride=1
1081 | pad=1
1082 | activation=leaky
1083 | 
1084 | [convolutional]
1085 | batch_normalize=1
1086 | filters=8
1087 | size=3
1088 | stride=1
1089 | pad=1
1090 | activation=leaky
1091 | 
1092 | [route]
1093 | layers=-1,-9
1094 | 
1095 | [convolutional]
1096 | batch_normalize=1
1097 | filters=72
1098 | size=1
1099 | stride=1
1100 | pad=1
1101 | activation=leaky
1102 | 
1103 | [convolutional]
1104 | size=1
1105 | stride=1
1106 | pad=1
1107 | filters=24
1108 | activation=linear
1109 | 
1110 | [yolo]
1111 | mask=6,7,8
1112 | anchors=40,39, 51,50, 61,59, 75,69, 62,92, 88,98, 115,77, 93,129, 128,115
1113 | classes=3
1114 | num=9
1115 | jitter=.3
1116 | ignore_thresh=.7
1117 | truth_thresh=1
1118 | scale_x_y=1.2
1119 | iou_thresh=0.213
1120 | cls_normalizer=1.0
1121 | iou_normalizer=0.07
1122 | iou_loss=ciou
1123 | nms_kind=greedynms
1124 | beta_nms=0.6
1125 | 
1126 | 


--------------------------------------------------------------------------------
/cfg/prune_0.8_keep_0.01_8x_yolov5l_v4.cfg:
--------------------------------------------------------------------------------
   1 | [net]
   2 | batch=64
   3 | subdivisions=8
   4 | width=416
   5 | height=416
   6 | channels=3
   7 | momentum=0.949
   8 | decay=0.0005
   9 | angle=0
  10 | saturation=1.5
  11 | exposure=1.5
  12 | hue=.1
  13 | learning_rate=0.00261
  14 | burn_in=1000
  15 | max_batches=500500
  16 | policy=steps
  17 | steps=400000,450000
  18 | scales=.1,.1
  19 | mosaic=1
  20 | 
  21 | [focus]
  22 | filters=12
  23 | 
  24 | [convolutional]
  25 | batch_normalize=1
  26 | filters=56
  27 | size=3
  28 | stride=1
  29 | pad=1
  30 | activation=leaky
  31 | 
  32 | [convolutional]
  33 | batch_normalize=1
  34 | filters=128
  35 | size=3
  36 | stride=2
  37 | pad=1
  38 | activation=leaky
  39 | 
  40 | [convolutional]
  41 | batch_normalize=1
  42 | filters=56
  43 | size=1
  44 | stride=1
  45 | pad=1
  46 | activation=leaky
  47 | 
  48 | [route]
  49 | layers=-2
  50 | 
  51 | [convolutional]
  52 | batch_normalize=1
  53 | filters=64
  54 | size=1
  55 | stride=1
  56 | pad=1
  57 | activation=leaky
  58 | 
  59 | [convolutional]
  60 | batch_normalize=1
  61 | filters=56
  62 | size=1
  63 | stride=1
  64 | pad=1
  65 | activation=leaky
  66 | 
  67 | [convolutional]
  68 | batch_normalize=1
  69 | filters=64
  70 | size=3
  71 | stride=1
  72 | pad=1
  73 | activation=leaky
  74 | 
  75 | [shortcut]
  76 | from=-3
  77 | activation=linear
  78 | 
  79 | [convolutional]
  80 | batch_normalize=1
  81 | filters=48
  82 | size=1
  83 | stride=1
  84 | pad=1
  85 | activation=leaky
  86 | 
  87 | [convolutional]
  88 | batch_normalize=1
  89 | filters=64
  90 | size=3
  91 | stride=1
  92 | pad=1
  93 | activation=leaky
  94 | 
  95 | [shortcut]
  96 | from=-3
  97 | activation=linear
  98 | 
  99 | [convolutional]
 100 | batch_normalize=1
 101 | filters=16
 102 | size=1
 103 | stride=1
 104 | pad=1
 105 | activation=leaky
 106 | 
 107 | [convolutional]
 108 | batch_normalize=1
 109 | filters=64
 110 | size=3
 111 | stride=1
 112 | pad=1
 113 | activation=leaky
 114 | 
 115 | [shortcut]
 116 | from=-3
 117 | activation=linear
 118 | 
 119 | [route]
 120 | layers=-1,-12
 121 | 
 122 | [convolutional]
 123 | batch_normalize=1
 124 | filters=64
 125 | size=1
 126 | stride=1
 127 | pad=1
 128 | activation=leaky
 129 | 
 130 | [convolutional]
 131 | batch_normalize=1
 132 | filters=16
 133 | size=3
 134 | stride=2
 135 | pad=1
 136 | activation=leaky
 137 | 
 138 | [convolutional]
 139 | batch_normalize=1
 140 | filters=128
 141 | size=1
 142 | stride=1
 143 | pad=1
 144 | activation=leaky
 145 | 
 146 | [route]
 147 | layers=-2
 148 | 
 149 | [convolutional]
 150 | batch_normalize=1
 151 | filters=109
 152 | size=1
 153 | stride=1
 154 | pad=1
 155 | activation=leaky
 156 | 
 157 | [convolutional]
 158 | batch_normalize=1
 159 | filters=32
 160 | size=1
 161 | stride=1
 162 | pad=1
 163 | activation=leaky
 164 | 
 165 | [convolutional]
 166 | batch_normalize=1
 167 | filters=109
 168 | size=3
 169 | stride=1
 170 | pad=1
 171 | activation=leaky
 172 | 
 173 | [shortcut]
 174 | from=-3
 175 | activation=linear
 176 | 
 177 | [convolutional]
 178 | batch_normalize=1
 179 | filters=16
 180 | size=1
 181 | stride=1
 182 | pad=1
 183 | activation=leaky
 184 | 
 185 | [convolutional]
 186 | batch_normalize=1
 187 | filters=109
 188 | size=3
 189 | stride=1
 190 | pad=1
 191 | activation=leaky
 192 | 
 193 | [shortcut]
 194 | from=-3
 195 | activation=linear
 196 | 
 197 | [convolutional]
 198 | batch_normalize=1
 199 | filters=8
 200 | size=1
 201 | stride=1
 202 | pad=1
 203 | activation=leaky
 204 | 
 205 | [convolutional]
 206 | batch_normalize=1
 207 | filters=109
 208 | size=3
 209 | stride=1
 210 | pad=1
 211 | activation=leaky
 212 | 
 213 | [shortcut]
 214 | from=-3
 215 | activation=linear
 216 | 
 217 | [convolutional]
 218 | batch_normalize=1
 219 | filters=8
 220 | size=1
 221 | stride=1
 222 | pad=1
 223 | activation=leaky
 224 | 
 225 | [convolutional]
 226 | batch_normalize=1
 227 | filters=109
 228 | size=3
 229 | stride=1
 230 | pad=1
 231 | activation=leaky
 232 | 
 233 | [shortcut]
 234 | from=-3
 235 | activation=linear
 236 | 
 237 | [convolutional]
 238 | batch_normalize=1
 239 | filters=8
 240 | size=1
 241 | stride=1
 242 | pad=1
 243 | activation=leaky
 244 | 
 245 | [convolutional]
 246 | batch_normalize=1
 247 | filters=109
 248 | size=3
 249 | stride=1
 250 | pad=1
 251 | activation=leaky
 252 | 
 253 | [shortcut]
 254 | from=-3
 255 | activation=linear
 256 | 
 257 | [convolutional]
 258 | batch_normalize=1
 259 | filters=8
 260 | size=1
 261 | stride=1
 262 | pad=1
 263 | activation=leaky
 264 | 
 265 | [convolutional]
 266 | batch_normalize=1
 267 | filters=109
 268 | size=3
 269 | stride=1
 270 | pad=1
 271 | activation=leaky
 272 | 
 273 | [shortcut]
 274 | from=-3
 275 | activation=linear
 276 | 
 277 | [convolutional]
 278 | batch_normalize=1
 279 | filters=8
 280 | size=1
 281 | stride=1
 282 | pad=1
 283 | activation=leaky
 284 | 
 285 | [convolutional]
 286 | batch_normalize=1
 287 | filters=109
 288 | size=3
 289 | stride=1
 290 | pad=1
 291 | activation=leaky
 292 | 
 293 | [shortcut]
 294 | from=-3
 295 | activation=linear
 296 | 
 297 | [convolutional]
 298 | batch_normalize=1
 299 | filters=8
 300 | size=1
 301 | stride=1
 302 | pad=1
 303 | activation=leaky
 304 | 
 305 | [convolutional]
 306 | batch_normalize=1
 307 | filters=109
 308 | size=3
 309 | stride=1
 310 | pad=1
 311 | activation=leaky
 312 | 
 313 | [shortcut]
 314 | from=-3
 315 | activation=linear
 316 | 
 317 | [convolutional]
 318 | batch_normalize=1
 319 | filters=8
 320 | size=1
 321 | stride=1
 322 | pad=1
 323 | activation=leaky
 324 | 
 325 | [convolutional]
 326 | batch_normalize=1
 327 | filters=109
 328 | size=3
 329 | stride=1
 330 | pad=1
 331 | activation=leaky
 332 | 
 333 | [shortcut]
 334 | from=-3
 335 | activation=linear
 336 | 
 337 | [route]
 338 | layers=-1,-30
 339 | 
 340 | [convolutional]
 341 | batch_normalize=1
 342 | filters=32
 343 | size=1
 344 | stride=1
 345 | pad=1
 346 | activation=leaky
 347 | 
 348 | [convolutional]
 349 | batch_normalize=1
 350 | filters=32
 351 | size=3
 352 | stride=2
 353 | pad=1
 354 | activation=leaky
 355 | 
 356 | [convolutional]
 357 | batch_normalize=1
 358 | filters=256
 359 | size=1
 360 | stride=1
 361 | pad=1
 362 | activation=leaky
 363 | 
 364 | [route]
 365 | layers=-2
 366 | 
 367 | [convolutional]
 368 | batch_normalize=1
 369 | filters=231
 370 | size=1
 371 | stride=1
 372 | pad=1
 373 | activation=leaky
 374 | 
 375 | [convolutional]
 376 | batch_normalize=1
 377 | filters=32
 378 | size=1
 379 | stride=1
 380 | pad=1
 381 | activation=leaky
 382 | 
 383 | [convolutional]
 384 | batch_normalize=1
 385 | filters=231
 386 | size=3
 387 | stride=1
 388 | pad=1
 389 | activation=leaky
 390 | 
 391 | [shortcut]
 392 | from=-3
 393 | activation=linear
 394 | 
 395 | [convolutional]
 396 | batch_normalize=1
 397 | filters=32
 398 | size=1
 399 | stride=1
 400 | pad=1
 401 | activation=leaky
 402 | 
 403 | [convolutional]
 404 | batch_normalize=1
 405 | filters=231
 406 | size=3
 407 | stride=1
 408 | pad=1
 409 | activation=leaky
 410 | 
 411 | [shortcut]
 412 | from=-3
 413 | activation=linear
 414 | 
 415 | [convolutional]
 416 | batch_normalize=1
 417 | filters=32
 418 | size=1
 419 | stride=1
 420 | pad=1
 421 | activation=leaky
 422 | 
 423 | [convolutional]
 424 | batch_normalize=1
 425 | filters=231
 426 | size=3
 427 | stride=1
 428 | pad=1
 429 | activation=leaky
 430 | 
 431 | [shortcut]
 432 | from=-3
 433 | activation=linear
 434 | 
 435 | [convolutional]
 436 | batch_normalize=1
 437 | filters=16
 438 | size=1
 439 | stride=1
 440 | pad=1
 441 | activation=leaky
 442 | 
 443 | [convolutional]
 444 | batch_normalize=1
 445 | filters=231
 446 | size=3
 447 | stride=1
 448 | pad=1
 449 | activation=leaky
 450 | 
 451 | [shortcut]
 452 | from=-3
 453 | activation=linear
 454 | 
 455 | [convolutional]
 456 | batch_normalize=1
 457 | filters=24
 458 | size=1
 459 | stride=1
 460 | pad=1
 461 | activation=leaky
 462 | 
 463 | [convolutional]
 464 | batch_normalize=1
 465 | filters=231
 466 | size=3
 467 | stride=1
 468 | pad=1
 469 | activation=leaky
 470 | 
 471 | [shortcut]
 472 | from=-3
 473 | activation=linear
 474 | 
 475 | [convolutional]
 476 | batch_normalize=1
 477 | filters=16
 478 | size=1
 479 | stride=1
 480 | pad=1
 481 | activation=leaky
 482 | 
 483 | [convolutional]
 484 | batch_normalize=1
 485 | filters=231
 486 | size=3
 487 | stride=1
 488 | pad=1
 489 | activation=leaky
 490 | 
 491 | [shortcut]
 492 | from=-3
 493 | activation=linear
 494 | 
 495 | [convolutional]
 496 | batch_normalize=1
 497 | filters=8
 498 | size=1
 499 | stride=1
 500 | pad=1
 501 | activation=leaky
 502 | 
 503 | [convolutional]
 504 | batch_normalize=1
 505 | filters=231
 506 | size=3
 507 | stride=1
 508 | pad=1
 509 | activation=leaky
 510 | 
 511 | [shortcut]
 512 | from=-3
 513 | activation=linear
 514 | 
 515 | [convolutional]
 516 | batch_normalize=1
 517 | filters=8
 518 | size=1
 519 | stride=1
 520 | pad=1
 521 | activation=leaky
 522 | 
 523 | [convolutional]
 524 | batch_normalize=1
 525 | filters=231
 526 | size=3
 527 | stride=1
 528 | pad=1
 529 | activation=leaky
 530 | 
 531 | [shortcut]
 532 | from=-3
 533 | activation=linear
 534 | 
 535 | [convolutional]
 536 | batch_normalize=1
 537 | filters=8
 538 | size=1
 539 | stride=1
 540 | pad=1
 541 | activation=leaky
 542 | 
 543 | [convolutional]
 544 | batch_normalize=1
 545 | filters=231
 546 | size=3
 547 | stride=1
 548 | pad=1
 549 | activation=leaky
 550 | 
 551 | [shortcut]
 552 | from=-3
 553 | activation=linear
 554 | 
 555 | [route]
 556 | layers=-1,-30
 557 | 
 558 | [convolutional]
 559 | batch_normalize=1
 560 | filters=96
 561 | size=1
 562 | stride=1
 563 | pad=1
 564 | activation=leaky
 565 | 
 566 | [convolutional]
 567 | batch_normalize=1
 568 | filters=408
 569 | size=3
 570 | stride=2
 571 | pad=1
 572 | activation=leaky
 573 | 
 574 | [convolutional]
 575 | batch_normalize=1
 576 | filters=512
 577 | size=1
 578 | stride=1
 579 | pad=1
 580 | activation=leaky
 581 | 
 582 | [maxpool]
 583 | stride=1
 584 | size=5
 585 | 
 586 | [route]
 587 | layers=-2
 588 | 
 589 | [maxpool]
 590 | stride=1
 591 | size=9
 592 | 
 593 | [route]
 594 | layers=-4
 595 | 
 596 | [maxpool]
 597 | stride=1
 598 | size=13
 599 | 
 600 | [route]
 601 | layers=-6,-5,-3,-1
 602 | 
 603 | [convolutional]
 604 | batch_normalize=1
 605 | filters=320
 606 | size=1
 607 | stride=1
 608 | pad=1
 609 | activation=leaky
 610 | 
 611 | [convolutional]
 612 | batch_normalize=1
 613 | filters=432
 614 | size=1
 615 | stride=1
 616 | pad=1
 617 | activation=leaky
 618 | 
 619 | [route]
 620 | layers=-2
 621 | 
 622 | [convolutional]
 623 | batch_normalize=1
 624 | filters=152
 625 | size=1
 626 | stride=1
 627 | pad=1
 628 | activation=leaky
 629 | 
 630 | [convolutional]
 631 | batch_normalize=1
 632 | filters=56
 633 | size=1
 634 | stride=1
 635 | pad=1
 636 | activation=leaky
 637 | 
 638 | [convolutional]
 639 | batch_normalize=1
 640 | filters=48
 641 | size=3
 642 | stride=1
 643 | pad=1
 644 | activation=leaky
 645 | 
 646 | [convolutional]
 647 | batch_normalize=1
 648 | filters=32
 649 | size=1
 650 | stride=1
 651 | pad=1
 652 | activation=leaky
 653 | 
 654 | [convolutional]
 655 | batch_normalize=1
 656 | filters=56
 657 | size=3
 658 | stride=1
 659 | pad=1
 660 | activation=leaky
 661 | 
 662 | [convolutional]
 663 | batch_normalize=1
 664 | filters=48
 665 | size=1
 666 | stride=1
 667 | pad=1
 668 | activation=leaky
 669 | 
 670 | [convolutional]
 671 | batch_normalize=1
 672 | filters=16
 673 | size=3
 674 | stride=1
 675 | pad=1
 676 | activation=leaky
 677 | 
 678 | [route]
 679 | layers=-1,-9
 680 | 
 681 | [convolutional]
 682 | batch_normalize=1
 683 | filters=464
 684 | size=1
 685 | stride=1
 686 | pad=1
 687 | activation=leaky
 688 | 
 689 | [convolutional]
 690 | batch_normalize=1
 691 | filters=512
 692 | size=1
 693 | stride=1
 694 | pad=1
 695 | activation=leaky
 696 | 
 697 | [upsample]
 698 | stride=2
 699 | 
 700 | [route]
 701 | layers=-1,-23
 702 | 
 703 | [convolutional]
 704 | batch_normalize=1
 705 | filters=96
 706 | size=1
 707 | stride=1
 708 | pad=1
 709 | activation=leaky
 710 | 
 711 | [route]
 712 | layers=-2
 713 | 
 714 | [convolutional]
 715 | batch_normalize=1
 716 | filters=56
 717 | size=1
 718 | stride=1
 719 | pad=1
 720 | activation=leaky
 721 | 
 722 | [convolutional]
 723 | batch_normalize=1
 724 | filters=80
 725 | size=1
 726 | stride=1
 727 | pad=1
 728 | activation=leaky
 729 | 
 730 | [convolutional]
 731 | batch_normalize=1
 732 | filters=24
 733 | size=3
 734 | stride=1
 735 | pad=1
 736 | activation=leaky
 737 | 
 738 | [convolutional]
 739 | batch_normalize=1
 740 | filters=64
 741 | size=1
 742 | stride=1
 743 | pad=1
 744 | activation=leaky
 745 | 
 746 | [convolutional]
 747 | batch_normalize=1
 748 | filters=40
 749 | size=3
 750 | stride=1
 751 | pad=1
 752 | activation=leaky
 753 | 
 754 | [convolutional]
 755 | batch_normalize=1
 756 | filters=48
 757 | size=1
 758 | stride=1
 759 | pad=1
 760 | activation=leaky
 761 | 
 762 | [convolutional]
 763 | batch_normalize=1
 764 | filters=32
 765 | size=3
 766 | stride=1
 767 | pad=1
 768 | activation=leaky
 769 | 
 770 | [route]
 771 | layers=-1,-9
 772 | 
 773 | [convolutional]
 774 | batch_normalize=1
 775 | filters=168
 776 | size=1
 777 | stride=1
 778 | pad=1
 779 | activation=leaky
 780 | 
 781 | [convolutional]
 782 | batch_normalize=1
 783 | filters=256
 784 | size=1
 785 | stride=1
 786 | pad=1
 787 | activation=leaky
 788 | 
 789 | [upsample]
 790 | stride=2
 791 | 
 792 | [route]
 793 | layers=-1,-70
 794 | 
 795 | [convolutional]
 796 | batch_normalize=1
 797 | filters=24
 798 | size=1
 799 | stride=1
 800 | pad=1
 801 | activation=leaky
 802 | 
 803 | [route]
 804 | layers=-2
 805 | 
 806 | [convolutional]
 807 | batch_normalize=1
 808 | filters=8
 809 | size=1
 810 | stride=1
 811 | pad=1
 812 | activation=leaky
 813 | 
 814 | [convolutional]
 815 | batch_normalize=1
 816 | filters=8
 817 | size=1
 818 | stride=1
 819 | pad=1
 820 | activation=leaky
 821 | 
 822 | [convolutional]
 823 | batch_normalize=1
 824 | filters=8
 825 | size=3
 826 | stride=1
 827 | pad=1
 828 | activation=leaky
 829 | 
 830 | [convolutional]
 831 | batch_normalize=1
 832 | filters=8
 833 | size=1
 834 | stride=1
 835 | pad=1
 836 | activation=leaky
 837 | 
 838 | [convolutional]
 839 | batch_normalize=1
 840 | filters=8
 841 | size=3
 842 | stride=1
 843 | pad=1
 844 | activation=leaky
 845 | 
 846 | [convolutional]
 847 | batch_normalize=1
 848 | filters=8
 849 | size=1
 850 | stride=1
 851 | pad=1
 852 | activation=leaky
 853 | 
 854 | [convolutional]
 855 | batch_normalize=1
 856 | filters=24
 857 | size=3
 858 | stride=1
 859 | pad=1
 860 | activation=leaky
 861 | 
 862 | [route]
 863 | layers=-1,-9
 864 | 
 865 | [convolutional]
 866 | batch_normalize=1
 867 | filters=176
 868 | size=1
 869 | stride=1
 870 | pad=1
 871 | activation=leaky
 872 | 
 873 | [convolutional]
 874 | size=1
 875 | stride=1
 876 | pad=1
 877 | filters=24
 878 | activation=linear
 879 | 
 880 | [yolo]
 881 | mask=0,1,2
 882 | anchors=39,38, 49,67, 74,49, 74,86, 113,71, 97,119, 163,108, 134,155, 210,199
 883 | classes=3
 884 | num=9
 885 | jitter=.3
 886 | ignore_thresh=.7
 887 | truth_thresh=1
 888 | scale_x_y=1.2
 889 | iou_thresh=0.213
 890 | cls_normalizer=1.0
 891 | iou_normalizer=0.07
 892 | iou_loss=ciou
 893 | nms_kind=greedynms
 894 | beta_nms=0.6
 895 | 
 896 | [route]
 897 | layers=-3
 898 | 
 899 | [convolutional]
 900 | batch_normalize=1
 901 | filters=32
 902 | size=3
 903 | stride=2
 904 | pad=1
 905 | activation=leaky
 906 | 
 907 | [route]
 908 | layers=-1,-18
 909 | 
 910 | [convolutional]
 911 | batch_normalize=1
 912 | filters=56
 913 | size=1
 914 | stride=1
 915 | pad=1
 916 | activation=leaky
 917 | 
 918 | [route]
 919 | layers=-2
 920 | 
 921 | [convolutional]
 922 | batch_normalize=1
 923 | filters=48
 924 | size=1
 925 | stride=1
 926 | pad=1
 927 | activation=leaky
 928 | 
 929 | [convolutional]
 930 | batch_normalize=1
 931 | filters=32
 932 | size=1
 933 | stride=1
 934 | pad=1
 935 | activation=leaky
 936 | 
 937 | [convolutional]
 938 | batch_normalize=1
 939 | filters=32
 940 | size=3
 941 | stride=1
 942 | pad=1
 943 | activation=leaky
 944 | 
 945 | [convolutional]
 946 | batch_normalize=1
 947 | filters=24
 948 | size=1
 949 | stride=1
 950 | pad=1
 951 | activation=leaky
 952 | 
 953 | [convolutional]
 954 | batch_normalize=1
 955 | filters=24
 956 | size=3
 957 | stride=1
 958 | pad=1
 959 | activation=leaky
 960 | 
 961 | [convolutional]
 962 | batch_normalize=1
 963 | filters=40
 964 | size=1
 965 | stride=1
 966 | pad=1
 967 | activation=leaky
 968 | 
 969 | [convolutional]
 970 | batch_normalize=1
 971 | filters=64
 972 | size=3
 973 | stride=1
 974 | pad=1
 975 | activation=leaky
 976 | 
 977 | [route]
 978 | layers=-1,-9
 979 | 
 980 | [convolutional]
 981 | batch_normalize=1
 982 | filters=240
 983 | size=1
 984 | stride=1
 985 | pad=1
 986 | activation=leaky
 987 | 
 988 | [convolutional]
 989 | size=1
 990 | stride=1
 991 | pad=1
 992 | filters=24
 993 | activation=linear
 994 | 
 995 | [yolo]
 996 | mask=3,4,5
 997 | anchors=39,38, 49,67, 74,49, 74,86, 113,71, 97,119, 163,108, 134,155, 210,199
 998 | classes=3
 999 | num=9
1000 | jitter=.3
1001 | ignore_thresh=.7
1002 | truth_thresh=1
1003 | scale_x_y=1.2
1004 | iou_thresh=0.213
1005 | cls_normalizer=1.0
1006 | iou_normalizer=0.07
1007 | iou_loss=ciou
1008 | nms_kind=greedynms
1009 | beta_nms=0.6
1010 | 
1011 | [route]
1012 | layers=-3
1013 | 
1014 | [convolutional]
1015 | batch_normalize=1
1016 | filters=176
1017 | size=3
1018 | stride=2
1019 | pad=1
1020 | activation=leaky
1021 | 
1022 | [route]
1023 | layers=-1,-48
1024 | 
1025 | [convolutional]
1026 | batch_normalize=1
1027 | filters=144
1028 | size=1
1029 | stride=1
1030 | pad=1
1031 | activation=leaky
1032 | 
1033 | [route]
1034 | layers=-2
1035 | 
1036 | [convolutional]
1037 | batch_normalize=1
1038 | filters=32
1039 | size=1
1040 | stride=1
1041 | pad=1
1042 | activation=leaky
1043 | 
1044 | [convolutional]
1045 | batch_normalize=1
1046 | filters=16
1047 | size=1
1048 | stride=1
1049 | pad=1
1050 | activation=leaky
1051 | 
1052 | [convolutional]
1053 | batch_normalize=1
1054 | filters=32
1055 | size=3
1056 | stride=1
1057 | pad=1
1058 | activation=leaky
1059 | 
1060 | [convolutional]
1061 | batch_normalize=1
1062 | filters=16
1063 | size=1
1064 | stride=1
1065 | pad=1
1066 | activation=leaky
1067 | 
1068 | [convolutional]
1069 | batch_normalize=1
1070 | filters=64
1071 | size=3
1072 | stride=1
1073 | pad=1
1074 | activation=leaky
1075 | 
1076 | [convolutional]
1077 | batch_normalize=1
1078 | filters=88
1079 | size=1
1080 | stride=1
1081 | pad=1
1082 | activation=leaky
1083 | 
1084 | [convolutional]
1085 | batch_normalize=1
1086 | filters=96
1087 | size=3
1088 | stride=1
1089 | pad=1
1090 | activation=leaky
1091 | 
1092 | [route]
1093 | layers=-1,-9
1094 | 
1095 | [convolutional]
1096 | batch_normalize=1
1097 | filters=272
1098 | size=1
1099 | stride=1
1100 | pad=1
1101 | activation=leaky
1102 | 
1103 | [convolutional]
1104 | size=1
1105 | stride=1
1106 | pad=1
1107 | filters=24
1108 | activation=linear
1109 | 
1110 | [yolo]
1111 | mask=6,7,8
1112 | anchors=39,38, 49,67, 74,49, 74,86, 113,71, 97,119, 163,108, 134,155, 210,199
1113 | classes=3
1114 | num=9
1115 | jitter=.3
1116 | ignore_thresh=.7
1117 | truth_thresh=1
1118 | scale_x_y=1.2
1119 | iou_thresh=0.213
1120 | cls_normalizer=1.0
1121 | iou_normalizer=0.07
1122 | iou_loss=ciou
1123 | nms_kind=greedynms
1124 | beta_nms=0.6
1125 | 
1126 | 


--------------------------------------------------------------------------------
/data/coco.data:
--------------------------------------------------------------------------------
1 | classes=80
2 | train=../coco/trainvalno5k.txt
3 | valid=../coco/5k.txt
4 | names=data/coco.names
5 | backup=backup/
6 | eval=coco
7 | 


--------------------------------------------------------------------------------
/data/coco.names:
--------------------------------------------------------------------------------
 1 | person
 2 | bicycle
 3 | car
 4 | motorcycle
 5 | airplane
 6 | bus
 7 | train
 8 | truck
 9 | boat
10 | traffic light
11 | fire hydrant
12 | stop sign
13 | parking meter
14 | bench
15 | bird
16 | cat
17 | dog
18 | horse
19 | sheep
20 | cow
21 | elephant
22 | bear
23 | zebra
24 | giraffe
25 | backpack
26 | umbrella
27 | handbag
28 | tie
29 | suitcase
30 | frisbee
31 | skis
32 | snowboard
33 | sports ball
34 | kite
35 | baseball bat
36 | baseball glove
37 | skateboard
38 | surfboard
39 | tennis racket
40 | bottle
41 | wine glass
42 | cup
43 | fork
44 | knife
45 | spoon
46 | bowl
47 | banana
48 | apple
49 | sandwich
50 | orange
51 | broccoli
52 | carrot
53 | hot dog
54 | pizza
55 | donut
56 | cake
57 | chair
58 | couch
59 | potted plant
60 | bed
61 | dining table
62 | toilet
63 | tv
64 | laptop
65 | mouse
66 | remote
67 | keyboard
68 | cell phone
69 | microwave
70 | oven
71 | toaster
72 | sink
73 | refrigerator
74 | book
75 | clock
76 | vase
77 | scissors
78 | teddy bear
79 | hair drier
80 | toothbrush
81 | 


--------------------------------------------------------------------------------
/data/coco_128img.data:
--------------------------------------------------------------------------------
1 | classes=80
2 | train=./data/coco_128img.txt
3 | valid=./data/coco_128img.txt
4 | names=data/coco.names
5 | backup=backup/
6 | eval=coco
7 | 


--------------------------------------------------------------------------------
/data/coco_128img.txt:
--------------------------------------------------------------------------------
  1 | /home/lishuang/Disk/remote/pycharm/yolov5_prune/data/coco128/images/train2017/000000000009.jpg
  2 | /home/lishuang/Disk/remote/pycharm/yolov5_prune/data/coco128/images/train2017/000000000025.jpg
  3 | /home/lishuang/Disk/remote/pycharm/yolov5_prune/data/coco128/images/train2017/000000000030.jpg
  4 | /home/lishuang/Disk/remote/pycharm/yolov5_prune/data/coco128/images/train2017/000000000034.jpg
  5 | /home/lishuang/Disk/remote/pycharm/yolov5_prune/data/coco128/images/train2017/000000000036.jpg
  6 | /home/lishuang/Disk/remote/pycharm/yolov5_prune/data/coco128/images/train2017/000000000042.jpg
  7 | /home/lishuang/Disk/remote/pycharm/yolov5_prune/data/coco128/images/train2017/000000000049.jpg
  8 | /home/lishuang/Disk/remote/pycharm/yolov5_prune/data/coco128/images/train2017/000000000061.jpg
  9 | /home/lishuang/Disk/remote/pycharm/yolov5_prune/data/coco128/images/train2017/000000000064.jpg
 10 | /home/lishuang/Disk/remote/pycharm/yolov5_prune/data/coco128/images/train2017/000000000071.jpg
 11 | /home/lishuang/Disk/remote/pycharm/yolov5_prune/data/coco128/images/train2017/000000000072.jpg
 12 | /home/lishuang/Disk/remote/pycharm/yolov5_prune/data/coco128/images/train2017/000000000073.jpg
 13 | /home/lishuang/Disk/remote/pycharm/yolov5_prune/data/coco128/images/train2017/000000000074.jpg
 14 | /home/lishuang/Disk/remote/pycharm/yolov5_prune/data/coco128/images/train2017/000000000077.jpg
 15 | /home/lishuang/Disk/remote/pycharm/yolov5_prune/data/coco128/images/train2017/000000000078.jpg
 16 | /home/lishuang/Disk/remote/pycharm/yolov5_prune/data/coco128/images/train2017/000000000081.jpg
 17 | /home/lishuang/Disk/remote/pycharm/yolov5_prune/data/coco128/images/train2017/000000000086.jpg
 18 | /home/lishuang/Disk/remote/pycharm/yolov5_prune/data/coco128/images/train2017/000000000089.jpg
 19 | /home/lishuang/Disk/remote/pycharm/yolov5_prune/data/coco128/images/train2017/000000000092.jpg
 20 | /home/lishuang/Disk/remote/pycharm/yolov5_prune/data/coco128/images/train2017/000000000094.jpg
 21 | /home/lishuang/Disk/remote/pycharm/yolov5_prune/data/coco128/images/train2017/000000000109.jpg
 22 | /home/lishuang/Disk/remote/pycharm/yolov5_prune/data/coco128/images/train2017/000000000110.jpg
 23 | /home/lishuang/Disk/remote/pycharm/yolov5_prune/data/coco128/images/train2017/000000000113.jpg
 24 | /home/lishuang/Disk/remote/pycharm/yolov5_prune/data/coco128/images/train2017/000000000127.jpg
 25 | /home/lishuang/Disk/remote/pycharm/yolov5_prune/data/coco128/images/train2017/000000000133.jpg
 26 | /home/lishuang/Disk/remote/pycharm/yolov5_prune/data/coco128/images/train2017/000000000136.jpg
 27 | /home/lishuang/Disk/remote/pycharm/yolov5_prune/data/coco128/images/train2017/000000000138.jpg
 28 | /home/lishuang/Disk/remote/pycharm/yolov5_prune/data/coco128/images/train2017/000000000142.jpg
 29 | /home/lishuang/Disk/remote/pycharm/yolov5_prune/data/coco128/images/train2017/000000000143.jpg
 30 | /home/lishuang/Disk/remote/pycharm/yolov5_prune/data/coco128/images/train2017/000000000144.jpg
 31 | /home/lishuang/Disk/remote/pycharm/yolov5_prune/data/coco128/images/train2017/000000000149.jpg
 32 | /home/lishuang/Disk/remote/pycharm/yolov5_prune/data/coco128/images/train2017/000000000151.jpg
 33 | /home/lishuang/Disk/remote/pycharm/yolov5_prune/data/coco128/images/train2017/000000000154.jpg
 34 | /home/lishuang/Disk/remote/pycharm/yolov5_prune/data/coco128/images/train2017/000000000164.jpg
 35 | /home/lishuang/Disk/remote/pycharm/yolov5_prune/data/coco128/images/train2017/000000000165.jpg
 36 | /home/lishuang/Disk/remote/pycharm/yolov5_prune/data/coco128/images/train2017/000000000192.jpg
 37 | /home/lishuang/Disk/remote/pycharm/yolov5_prune/data/coco128/images/train2017/000000000194.jpg
 38 | /home/lishuang/Disk/remote/pycharm/yolov5_prune/data/coco128/images/train2017/000000000196.jpg
 39 | /home/lishuang/Disk/remote/pycharm/yolov5_prune/data/coco128/images/train2017/000000000201.jpg
 40 | /home/lishuang/Disk/remote/pycharm/yolov5_prune/data/coco128/images/train2017/000000000208.jpg
 41 | /home/lishuang/Disk/remote/pycharm/yolov5_prune/data/coco128/images/train2017/000000000241.jpg
 42 | /home/lishuang/Disk/remote/pycharm/yolov5_prune/data/coco128/images/train2017/000000000247.jpg
 43 | /home/lishuang/Disk/remote/pycharm/yolov5_prune/data/coco128/images/train2017/000000000250.jpg
 44 | /home/lishuang/Disk/remote/pycharm/yolov5_prune/data/coco128/images/train2017/000000000257.jpg
 45 | /home/lishuang/Disk/remote/pycharm/yolov5_prune/data/coco128/images/train2017/000000000260.jpg
 46 | /home/lishuang/Disk/remote/pycharm/yolov5_prune/data/coco128/images/train2017/000000000263.jpg
 47 | /home/lishuang/Disk/remote/pycharm/yolov5_prune/data/coco128/images/train2017/000000000283.jpg
 48 | /home/lishuang/Disk/remote/pycharm/yolov5_prune/data/coco128/images/train2017/000000000294.jpg
 49 | /home/lishuang/Disk/remote/pycharm/yolov5_prune/data/coco128/images/train2017/000000000307.jpg
 50 | /home/lishuang/Disk/remote/pycharm/yolov5_prune/data/coco128/images/train2017/000000000308.jpg
 51 | /home/lishuang/Disk/remote/pycharm/yolov5_prune/data/coco128/images/train2017/000000000309.jpg
 52 | /home/lishuang/Disk/remote/pycharm/yolov5_prune/data/coco128/images/train2017/000000000312.jpg
 53 | /home/lishuang/Disk/remote/pycharm/yolov5_prune/data/coco128/images/train2017/000000000315.jpg
 54 | /home/lishuang/Disk/remote/pycharm/yolov5_prune/data/coco128/images/train2017/000000000321.jpg
 55 | /home/lishuang/Disk/remote/pycharm/yolov5_prune/data/coco128/images/train2017/000000000322.jpg
 56 | /home/lishuang/Disk/remote/pycharm/yolov5_prune/data/coco128/images/train2017/000000000326.jpg
 57 | /home/lishuang/Disk/remote/pycharm/yolov5_prune/data/coco128/images/train2017/000000000328.jpg
 58 | /home/lishuang/Disk/remote/pycharm/yolov5_prune/data/coco128/images/train2017/000000000332.jpg
 59 | /home/lishuang/Disk/remote/pycharm/yolov5_prune/data/coco128/images/train2017/000000000338.jpg
 60 | /home/lishuang/Disk/remote/pycharm/yolov5_prune/data/coco128/images/train2017/000000000349.jpg
 61 | /home/lishuang/Disk/remote/pycharm/yolov5_prune/data/coco128/images/train2017/000000000357.jpg
 62 | /home/lishuang/Disk/remote/pycharm/yolov5_prune/data/coco128/images/train2017/000000000359.jpg
 63 | /home/lishuang/Disk/remote/pycharm/yolov5_prune/data/coco128/images/train2017/000000000360.jpg
 64 | /home/lishuang/Disk/remote/pycharm/yolov5_prune/data/coco128/images/train2017/000000000368.jpg
 65 | /home/lishuang/Disk/remote/pycharm/yolov5_prune/data/coco128/images/train2017/000000000370.jpg
 66 | /home/lishuang/Disk/remote/pycharm/yolov5_prune/data/coco128/images/train2017/000000000382.jpg
 67 | /home/lishuang/Disk/remote/pycharm/yolov5_prune/data/coco128/images/train2017/000000000384.jpg
 68 | /home/lishuang/Disk/remote/pycharm/yolov5_prune/data/coco128/images/train2017/000000000387.jpg
 69 | /home/lishuang/Disk/remote/pycharm/yolov5_prune/data/coco128/images/train2017/000000000389.jpg
 70 | /home/lishuang/Disk/remote/pycharm/yolov5_prune/data/coco128/images/train2017/000000000394.jpg
 71 | /home/lishuang/Disk/remote/pycharm/yolov5_prune/data/coco128/images/train2017/000000000395.jpg
 72 | /home/lishuang/Disk/remote/pycharm/yolov5_prune/data/coco128/images/train2017/000000000397.jpg
 73 | /home/lishuang/Disk/remote/pycharm/yolov5_prune/data/coco128/images/train2017/000000000400.jpg
 74 | /home/lishuang/Disk/remote/pycharm/yolov5_prune/data/coco128/images/train2017/000000000404.jpg
 75 | /home/lishuang/Disk/remote/pycharm/yolov5_prune/data/coco128/images/train2017/000000000415.jpg
 76 | /home/lishuang/Disk/remote/pycharm/yolov5_prune/data/coco128/images/train2017/000000000419.jpg
 77 | /home/lishuang/Disk/remote/pycharm/yolov5_prune/data/coco128/images/train2017/000000000428.jpg
 78 | /home/lishuang/Disk/remote/pycharm/yolov5_prune/data/coco128/images/train2017/000000000431.jpg
 79 | /home/lishuang/Disk/remote/pycharm/yolov5_prune/data/coco128/images/train2017/000000000436.jpg
 80 | /home/lishuang/Disk/remote/pycharm/yolov5_prune/data/coco128/images/train2017/000000000438.jpg
 81 | /home/lishuang/Disk/remote/pycharm/yolov5_prune/data/coco128/images/train2017/000000000443.jpg
 82 | /home/lishuang/Disk/remote/pycharm/yolov5_prune/data/coco128/images/train2017/000000000446.jpg
 83 | /home/lishuang/Disk/remote/pycharm/yolov5_prune/data/coco128/images/train2017/000000000450.jpg
 84 | /home/lishuang/Disk/remote/pycharm/yolov5_prune/data/coco128/images/train2017/000000000459.jpg
 85 | /home/lishuang/Disk/remote/pycharm/yolov5_prune/data/coco128/images/train2017/000000000471.jpg
 86 | /home/lishuang/Disk/remote/pycharm/yolov5_prune/data/coco128/images/train2017/000000000472.jpg
 87 | /home/lishuang/Disk/remote/pycharm/yolov5_prune/data/coco128/images/train2017/000000000474.jpg
 88 | /home/lishuang/Disk/remote/pycharm/yolov5_prune/data/coco128/images/train2017/000000000486.jpg
 89 | /home/lishuang/Disk/remote/pycharm/yolov5_prune/data/coco128/images/train2017/000000000488.jpg
 90 | /home/lishuang/Disk/remote/pycharm/yolov5_prune/data/coco128/images/train2017/000000000490.jpg
 91 | /home/lishuang/Disk/remote/pycharm/yolov5_prune/data/coco128/images/train2017/000000000491.jpg
 92 | /home/lishuang/Disk/remote/pycharm/yolov5_prune/data/coco128/images/train2017/000000000502.jpg
 93 | /home/lishuang/Disk/remote/pycharm/yolov5_prune/data/coco128/images/train2017/000000000508.jpg
 94 | /home/lishuang/Disk/remote/pycharm/yolov5_prune/data/coco128/images/train2017/000000000510.jpg
 95 | /home/lishuang/Disk/remote/pycharm/yolov5_prune/data/coco128/images/train2017/000000000514.jpg
 96 | /home/lishuang/Disk/remote/pycharm/yolov5_prune/data/coco128/images/train2017/000000000520.jpg
 97 | /home/lishuang/Disk/remote/pycharm/yolov5_prune/data/coco128/images/train2017/000000000529.jpg
 98 | /home/lishuang/Disk/remote/pycharm/yolov5_prune/data/coco128/images/train2017/000000000531.jpg
 99 | /home/lishuang/Disk/remote/pycharm/yolov5_prune/data/coco128/images/train2017/000000000532.jpg
100 | /home/lishuang/Disk/remote/pycharm/yolov5_prune/data/coco128/images/train2017/000000000536.jpg
101 | /home/lishuang/Disk/remote/pycharm/yolov5_prune/data/coco128/images/train2017/000000000540.jpg
102 | /home/lishuang/Disk/remote/pycharm/yolov5_prune/data/coco128/images/train2017/000000000542.jpg
103 | /home/lishuang/Disk/remote/pycharm/yolov5_prune/data/coco128/images/train2017/000000000544.jpg
104 | /home/lishuang/Disk/remote/pycharm/yolov5_prune/data/coco128/images/train2017/000000000560.jpg
105 | /home/lishuang/Disk/remote/pycharm/yolov5_prune/data/coco128/images/train2017/000000000562.jpg
106 | /home/lishuang/Disk/remote/pycharm/yolov5_prune/data/coco128/images/train2017/000000000564.jpg
107 | /home/lishuang/Disk/remote/pycharm/yolov5_prune/data/coco128/images/train2017/000000000569.jpg
108 | /home/lishuang/Disk/remote/pycharm/yolov5_prune/data/coco128/images/train2017/000000000572.jpg
109 | /home/lishuang/Disk/remote/pycharm/yolov5_prune/data/coco128/images/train2017/000000000575.jpg
110 | /home/lishuang/Disk/remote/pycharm/yolov5_prune/data/coco128/images/train2017/000000000581.jpg
111 | /home/lishuang/Disk/remote/pycharm/yolov5_prune/data/coco128/images/train2017/000000000584.jpg
112 | /home/lishuang/Disk/remote/pycharm/yolov5_prune/data/coco128/images/train2017/000000000589.jpg
113 | /home/lishuang/Disk/remote/pycharm/yolov5_prune/data/coco128/images/train2017/000000000590.jpg
114 | /home/lishuang/Disk/remote/pycharm/yolov5_prune/data/coco128/images/train2017/000000000595.jpg
115 | /home/lishuang/Disk/remote/pycharm/yolov5_prune/data/coco128/images/train2017/000000000597.jpg
116 | /home/lishuang/Disk/remote/pycharm/yolov5_prune/data/coco128/images/train2017/000000000599.jpg
117 | /home/lishuang/Disk/remote/pycharm/yolov5_prune/data/coco128/images/train2017/000000000605.jpg
118 | /home/lishuang/Disk/remote/pycharm/yolov5_prune/data/coco128/images/train2017/000000000612.jpg
119 | /home/lishuang/Disk/remote/pycharm/yolov5_prune/data/coco128/images/train2017/000000000620.jpg
120 | /home/lishuang/Disk/remote/pycharm/yolov5_prune/data/coco128/images/train2017/000000000623.jpg
121 | /home/lishuang/Disk/remote/pycharm/yolov5_prune/data/coco128/images/train2017/000000000625.jpg
122 | /home/lishuang/Disk/remote/pycharm/yolov5_prune/data/coco128/images/train2017/000000000626.jpg
123 | /home/lishuang/Disk/remote/pycharm/yolov5_prune/data/coco128/images/train2017/000000000629.jpg
124 | /home/lishuang/Disk/remote/pycharm/yolov5_prune/data/coco128/images/train2017/000000000634.jpg
125 | /home/lishuang/Disk/remote/pycharm/yolov5_prune/data/coco128/images/train2017/000000000636.jpg
126 | /home/lishuang/Disk/remote/pycharm/yolov5_prune/data/coco128/images/train2017/000000000641.jpg
127 | /home/lishuang/Disk/remote/pycharm/yolov5_prune/data/coco128/images/train2017/000000000643.jpg
128 | /home/lishuang/Disk/remote/pycharm/yolov5_prune/data/coco128/images/train2017/000000000650.jpg


--------------------------------------------------------------------------------
/data/get_coco_dataset.sh:
--------------------------------------------------------------------------------
 1 | #!/bin/bash
 2 | # CREDIT: https://github.com/pjreddie/darknet/tree/master/scripts/get_coco_dataset.sh
 3 | 
 4 | # Clone COCO API
 5 | git clone https://github.com/pdollar/coco && cd coco
 6 | 
 7 | # Download Images
 8 | mkdir images && cd images
 9 | wget -c https://pjreddie.com/media/files/train2014.zip
10 | wget -c https://pjreddie.com/media/files/val2014.zip
11 | 
12 | # Unzip
13 | unzip -q train2014.zip
14 | unzip -q val2014.zip
15 | 
16 | # (optional) Delete zip files
17 | rm -rf *.zip
18 | 
19 | cd ..
20 | 
21 | # Download COCO Metadata
22 | wget -c https://pjreddie.com/media/files/instances_train-val2014.zip
23 | wget -c https://pjreddie.com/media/files/coco/5k.part
24 | wget -c https://pjreddie.com/media/files/coco/trainvalno5k.part
25 | wget -c https://pjreddie.com/media/files/coco/labels.tgz
26 | tar xzf labels.tgz
27 | unzip -q instances_train-val2014.zip
28 | 
29 | # Set Up Image Lists
30 | paste <(awk "{print \"$PWD\"}" <5k.part) 5k.part | tr -d '\t' > 5k.txt
31 | paste <(awk "{print \"$PWD\"}" <trainvalno5k.part) trainvalno5k.part | tr -d '\t' > trainvalno5k.txt
32 | 
33 | # get xview training data
34 | # wget -O train_images.tgz 'https://d307kc0mrhucc3.cloudfront.net/train_images.tgz?Expires=1530124049&Signature=JrQoxipmsETvb7eQHCfDFUO-QEHJGAayUv0i-ParmS-1hn7hl9D~bzGuHWG82imEbZSLUARTtm0wOJ7EmYMGmG5PtLKz9H5qi6DjoSUuFc13NQ-~6yUhE~NfPaTnehUdUMCa3On2wl1h1ZtRG~0Jq1P-AJbpe~oQxbyBrs1KccaMa7FK4F4oMM6sMnNgoXx8-3O77kYw~uOpTMFmTaQdHln6EztW0Lx17i57kK3ogbSUpXgaUTqjHCRA1dWIl7PY1ngQnLslkLhZqmKcaL-BvWf0ZGjHxCDQBpnUjIlvMu5NasegkwD9Jjc0ClgTxsttSkmbapVqaVC8peR0pO619Q__&Key-Pair-Id=APKAIKGDJB5C3XUL2DXQ'
35 | # tar -xvzf train_images.tgz
36 | # sudo rm -rf train_images/._*
37 | # lastly convert each .tif to a .bmp for faster loading in cv2
38 | 
39 | # ./coco/images/train2014/COCO_train2014_000000167126.jpg  # corrupted image
40 | 


--------------------------------------------------------------------------------
/data/get_coco_dataset_gdrive.sh:
--------------------------------------------------------------------------------
 1 | #!/bin/bash
 2 | # https://stackoverflow.com/questions/48133080/how-to-download-a-google-drive-url-via-curl-or-wget/48133859
 3 | 
 4 | # Zip coco folder
 5 | # zip -r coco.zip coco
 6 | # tar -czvf coco.tar.gz coco
 7 | 
 8 | # Set fileid and filename
 9 | filename="coco.zip"
10 | fileid="1HaXkef9z6y5l4vUnCYgdmEAj61c6bfWO"  # coco.zip
11 | 
12 | # Download from Google Drive, accepting presented query
13 | curl -c ./cookie -s -L "https://drive.google.com/uc?export=download&id=${fileid}" > /dev/null
14 | curl -Lb ./cookie "https://drive.google.com/uc?export=download&confirm=`awk '/download/ {print $NF}' ./cookie`&id=${fileid}" -o ${filename}
15 | rm ./cookie
16 | 
17 | # Unzip
18 | unzip -q ${filename}  # for coco.zip
19 | # tar -xzf ${filename}  # for coco.tar.gz
20 | 


--------------------------------------------------------------------------------
/models/__init__.py:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/BokyLiu/YoloV5sl_V4_prune/c0ff39c5a5b10cbca95beb597c722cdc02e81885/models/__init__.py


--------------------------------------------------------------------------------
/models/common.py:
--------------------------------------------------------------------------------
  1 | # This file contains modules common to various models
  2 | import math
  3 | 
  4 | import torch
  5 | import torch.nn as nn
  6 | 
  7 | 
  8 | def autopad(k, p=None):  # kernel, padding
  9 |     # Pad to 'same'
 10 |     if p is None:
 11 |         p = k // 2 if isinstance(k, int) else [x // 2 for x in k]  # auto-pad
 12 |     return p
 13 | 
 14 | 
 15 | def DWConv(c1, c2, k=1, s=1, act=True):
 16 |     # Depthwise convolution
 17 |     return Conv(c1, c2, k, s, g=math.gcd(c1, c2), act=act)
 18 | 
 19 | 
 20 | class Conv(nn.Module):
 21 |     # Standard convolution
 22 |     def __init__(self, c1, c2, k=1, s=1, p=None, g=1, act=True):  # ch_in, ch_out, kernel, stride, padding, groups
 23 |         super(Conv, self).__init__()
 24 |         self.conv = nn.Conv2d(c1, c2, k, s, autopad(k, p), groups=g, bias=False)
 25 |         self.bn = nn.BatchNorm2d(c2)
 26 |         #self.act = nn.LeakyReLU(0.1, inplace=True) if act else nn.Identity() #yolov5_v2
 27 |         #self.act = nn.Hardswish() if act else nn.Identity() #yolov5_v3
 28 |         self.act = nn.SiLU() if act is True else (act if isinstance(act, nn.Module) else nn.Identity()) #yolov5_v4
 29 | 
 30 |     def forward(self, x):
 31 |         return self.act(self.bn(self.conv(x)))
 32 | 
 33 |     def fuseforward(self, x):
 34 |         return self.act(self.conv(x))
 35 | 
 36 | 
 37 | class Bottleneck(nn.Module):
 38 |     # Standard bottleneck
 39 |     def __init__(self, c1, c2, shortcut=True, g=1, e=0.5):  # ch_in, ch_out, shortcut, groups, expansion
 40 |         super(Bottleneck, self).__init__()
 41 |         c_ = int(c2 * e)  # hidden channels
 42 |         self.cv1 = Conv(c1, c_, 1, 1)
 43 |         self.cv2 = Conv(c_, c2, 3, 1, g=g)
 44 |         self.add = shortcut and c1 == c2
 45 | 
 46 |     def forward(self, x):
 47 |         return x + self.cv2(self.cv1(x)) if self.add else self.cv2(self.cv1(x))
 48 | 
 49 | 
 50 | class BottleneckCSP(nn.Module):
 51 |     # CSP Bottleneck https://github.com/WongKinYiu/CrossStagePartialNetworks
 52 |     def __init__(self, c1, c2, n=1, shortcut=True, g=1, e=0.5):  # ch_in, ch_out, number, shortcut, groups, expansion
 53 |         super(BottleneckCSP, self).__init__()
 54 |         c_ = int(c2 * e)  # hidden channels
 55 |         self.cv1 = Conv(c1, c_, 1, 1)
 56 |         self.cv2 = nn.Conv2d(c1, c_, 1, 1, bias=False)
 57 |         self.cv3 = nn.Conv2d(c_, c_, 1, 1, bias=False)
 58 |         self.cv4 = Conv(2 * c_, c2, 1, 1)
 59 |         self.bn = nn.BatchNorm2d(2 * c_)  # applied to cat(cv2, cv3)
 60 |         self.act = nn.LeakyReLU(0.1, inplace=True)
 61 |         self.m = nn.Sequential(*[Bottleneck(c_, c_, shortcut, g, e=1.0) for _ in range(n)])
 62 | 
 63 |     def forward(self, x):
 64 |         y1 = self.cv3(self.m(self.cv1(x)))
 65 |         y2 = self.cv2(x)
 66 |         return self.cv4(self.act(self.bn(torch.cat((y1, y2), dim=1))))
 67 | 
 68 | class C3(nn.Module):
 69 |     # CSP Bottleneck with 3 convolutions
 70 |     def __init__(self, c1, c2, n=1, shortcut=True, g=1, e=0.5):  # ch_in, ch_out, number, shortcut, groups, expansion
 71 |         super(C3, self).__init__()
 72 |         c_ = int(c2 * e)  # hidden channels
 73 |         self.cv1 = Conv(c1, c_, 1, 1)
 74 |         self.cv2 = Conv(c1, c_, 1, 1)
 75 |         self.cv3 = Conv(2 * c_, c2, 1)  # act=FReLU(c2)
 76 |         self.m = nn.Sequential(*[Bottleneck(c_, c_, shortcut, g, e=1.0) for _ in range(n)])
 77 |         # self.m = nn.Sequential(*[CrossConv(c_, c_, 3, 1, g, 1.0, shortcut) for _ in range(n)])
 78 | 
 79 |     def forward(self, x):
 80 |         return self.cv3(torch.cat((self.m(self.cv1(x)), self.cv2(x)), dim=1))
 81 | 
 82 | class SPP(nn.Module):
 83 |     # Spatial pyramid pooling layer used in YOLOv3-SPP
 84 |     def __init__(self, c1, c2, k=(5, 9, 13)):
 85 |         super(SPP, self).__init__()
 86 |         c_ = c1 // 2  # hidden channels
 87 |         self.cv1 = Conv(c1, c_, 1, 1)
 88 |         self.cv2 = Conv(c_ * (len(k) + 1), c2, 1, 1)
 89 |         self.m = nn.ModuleList([nn.MaxPool2d(kernel_size=x, stride=1, padding=x // 2) for x in k])
 90 | 
 91 |     def forward(self, x):
 92 |         x = self.cv1(x)
 93 |         return self.cv2(torch.cat([x] + [m(x) for m in self.m], 1))
 94 | 
 95 | 
 96 | class Focus(nn.Module):
 97 |     # Focus wh information into c-space
 98 |     def __init__(self, c1, c2, k=1, s=1, p=None, g=1, act=True):  # ch_in, ch_out, kernel, stride, padding, groups
 99 |         super(Focus, self).__init__()
100 |         self.conv = Conv(c1 * 4, c2, k, s, p, g, act)
101 | 
102 |     def forward(self, x):  # x(b,c,w,h) -> y(b,4c,w/2,h/2)
103 |         return self.conv(torch.cat([x[..., ::2, ::2], x[..., 1::2, ::2], x[..., ::2, 1::2], x[..., 1::2, 1::2]], 1))
104 | 
105 | 
106 | class Concat(nn.Module):
107 |     # Concatenate a list of tensors along dimension
108 |     def __init__(self, dimension=1):
109 |         super(Concat, self).__init__()
110 |         self.d = dimension
111 | 
112 |     def forward(self, x):
113 |         return torch.cat(x, self.d)
114 | 
115 | 
116 | class Flatten(nn.Module):
117 |     # Use after nn.AdaptiveAvgPool2d(1) to remove last 2 dimensions
118 |     @staticmethod
119 |     def forward(x):
120 |         return x.view(x.size(0), -1)
121 | 
122 | 
123 | class Classify(nn.Module):
124 |     # Classification head, i.e. x(b,c1,20,20) to x(b,c2)
125 |     def __init__(self, c1, c2, k=1, s=1, p=None, g=1):  # ch_in, ch_out, kernel, stride, padding, groups
126 |         super(Classify, self).__init__()
127 |         self.aap = nn.AdaptiveAvgPool2d(1)  # to x(b,c1,1,1)
128 |         self.conv = nn.Conv2d(c1, c2, k, s, autopad(k, p), groups=g, bias=False)  # to x(b,c2,1,1)
129 |         self.flat = Flatten()
130 | 
131 |     def forward(self, x):
132 |         z = torch.cat([self.aap(y) for y in (x if isinstance(x, list) else [x])], 1)  # cat if list
133 |         return self.flat(self.conv(z))  # flatten to x(b,c2)
134 | 


--------------------------------------------------------------------------------
/models/experimental.py:
--------------------------------------------------------------------------------
  1 | # This file contains experimental modules
  2 | 
  3 | import numpy as np
  4 | import torch
  5 | import torch.nn as nn
  6 | 
  7 | from models.common import Conv, DWConv
  8 | # from utils.google_utils import attempt_download
  9 | 
 10 | 
 11 | class CrossConv(nn.Module):
 12 |     # Cross Convolution Downsample
 13 |     def __init__(self, c1, c2, k=3, s=1, g=1, e=1.0, shortcut=False):
 14 |         # ch_in, ch_out, kernel, stride, groups, expansion, shortcut
 15 |         super(CrossConv, self).__init__()
 16 |         c_ = int(c2 * e)  # hidden channels
 17 |         self.cv1 = Conv(c1, c_, (1, k), (1, s))
 18 |         self.cv2 = Conv(c_, c2, (k, 1), (s, 1), g=g)
 19 |         self.add = shortcut and c1 == c2
 20 | 
 21 |     def forward(self, x):
 22 |         return x + self.cv2(self.cv1(x)) if self.add else self.cv2(self.cv1(x))
 23 | 
 24 | class Sum(nn.Module):
 25 |     # Weighted sum of 2 or more layers https://arxiv.org/abs/1911.09070
 26 |     def __init__(self, n, weight=False):  # n: number of inputs
 27 |         super(Sum, self).__init__()
 28 |         self.weight = weight  # apply weights boolean
 29 |         self.iter = range(n - 1)  # iter object
 30 |         if weight:
 31 |             self.w = nn.Parameter(-torch.arange(1., n) / 2, requires_grad=True)  # layer weights
 32 | 
 33 |     def forward(self, x):
 34 |         y = x[0]  # no weight
 35 |         if self.weight:
 36 |             w = torch.sigmoid(self.w) * 2
 37 |             for i in self.iter:
 38 |                 y = y + x[i + 1] * w[i]
 39 |         else:
 40 |             for i in self.iter:
 41 |                 y = y + x[i + 1]
 42 |         return y
 43 | 
 44 | 
 45 | class GhostConv(nn.Module):
 46 |     # Ghost Convolution https://github.com/huawei-noah/ghostnet
 47 |     def __init__(self, c1, c2, k=1, s=1, g=1, act=True):  # ch_in, ch_out, kernel, stride, groups
 48 |         super(GhostConv, self).__init__()
 49 |         c_ = c2 // 2  # hidden channels
 50 |         self.cv1 = Conv(c1, c_, k, s, g, act)
 51 |         self.cv2 = Conv(c_, c_, 5, 1, c_, act)
 52 | 
 53 |     def forward(self, x):
 54 |         y = self.cv1(x)
 55 |         return torch.cat([y, self.cv2(y)], 1)
 56 | 
 57 | 
 58 | class GhostBottleneck(nn.Module):
 59 |     # Ghost Bottleneck https://github.com/huawei-noah/ghostnet
 60 |     def __init__(self, c1, c2, k, s):
 61 |         super(GhostBottleneck, self).__init__()
 62 |         c_ = c2 // 2
 63 |         self.conv = nn.Sequential(GhostConv(c1, c_, 1, 1),  # pw
 64 |                                   DWConv(c_, c_, k, s, act=False) if s == 2 else nn.Identity(),  # dw
 65 |                                   GhostConv(c_, c2, 1, 1, act=False))  # pw-linear
 66 |         self.shortcut = nn.Sequential(DWConv(c1, c1, k, s, act=False),
 67 |                                       Conv(c1, c2, 1, 1, act=False)) if s == 2 else nn.Identity()
 68 | 
 69 |     def forward(self, x):
 70 |         return self.conv(x) + self.shortcut(x)
 71 | 
 72 | 
 73 | class MixConv2d(nn.Module):
 74 |     # Mixed Depthwise Conv https://arxiv.org/abs/1907.09595
 75 |     def __init__(self, c1, c2, k=(1, 3), s=1, equal_ch=True):
 76 |         super(MixConv2d, self).__init__()
 77 |         groups = len(k)
 78 |         if equal_ch:  # equal c_ per group
 79 |             i = torch.linspace(0, groups - 1E-6, c2).floor()  # c2 indices
 80 |             c_ = [(i == g).sum() for g in range(groups)]  # intermediate channels
 81 |         else:  # equal weight.numel() per group
 82 |             b = [c2] + [0] * groups
 83 |             a = np.eye(groups + 1, groups, k=-1)
 84 |             a -= np.roll(a, 1, axis=1)
 85 |             a *= np.array(k) ** 2
 86 |             a[0] = 1
 87 |             c_ = np.linalg.lstsq(a, b, rcond=None)[0].round()  # solve for equal weight indices, ax = b
 88 | 
 89 |         self.m = nn.ModuleList([nn.Conv2d(c1, int(c_[g]), k[g], s, k[g] // 2, bias=False) for g in range(groups)])
 90 |         self.bn = nn.BatchNorm2d(c2)
 91 |         self.act = nn.LeakyReLU(0.1, inplace=True)
 92 | 
 93 |     def forward(self, x):
 94 |         return x + self.act(self.bn(torch.cat([m(x) for m in self.m], 1)))
 95 | 
 96 | 
 97 | class Ensemble(nn.ModuleList):
 98 |     # Ensemble of models
 99 |     def __init__(self):
100 |         super(Ensemble, self).__init__()
101 | 
102 |     def forward(self, x, augment=False):
103 |         y = []
104 |         for module in self:
105 |             y.append(module(x, augment)[0])
106 |         # y = torch.stack(y).max(0)[0]  # max ensemble
107 |         # y = torch.cat(y, 1)  # nms ensemble
108 |         y = torch.stack(y).mean(0)  # mean ensemble
109 |         return y, None  # inference, train output
110 | 
111 | 
112 | # def attempt_load(weights, map_location=None):
113 | #     # Loads an ensemble of models weights=[a,b,c] or a single model weights=[a] or weights=a
114 | #     model = Ensemble()
115 | #     for w in weights if isinstance(weights, list) else [weights]:
116 | #         attempt_download(w)
117 | #         model.append(torch.load(w, map_location=map_location)['model'].float().fuse().eval())  # load FP32 model
118 | #
119 | #     if len(model) == 1:
120 | #         return model[-1]  # return model
121 | #     else:
122 | #         print('Ensemble created with %s\n' % weights)
123 | #         for k in ['names', 'stride']:
124 | #             setattr(model, k, getattr(model[-1], k))
125 | #         return model  # return ensemble
126 | 


--------------------------------------------------------------------------------
/models/export.py:
--------------------------------------------------------------------------------
 1 | """Exports a YOLOv5 *.pt model to ONNX and TorchScript formats
 2 | 
 3 | Usage:
 4 |     $ export PYTHONPATH="$PWD" && python models/export.py --weights ./weights/yolov5s.pt --img 640 --batch 1
 5 | """
 6 | 
 7 | import argparse
 8 | 
 9 | import torch
10 | 
11 | from utils.google_utils import attempt_download
12 | 
13 | if __name__ == '__main__':
14 |     parser = argparse.ArgumentParser()
15 |     parser.add_argument('--weights', type=str, default='./yolov5s.pt', help='weights path')
16 |     parser.add_argument('--img-size', nargs='+', type=int, default=[640, 640], help='image size')
17 |     parser.add_argument('--batch-size', type=int, default=1, help='batch size')
18 |     opt = parser.parse_args()
19 |     opt.img_size *= 2 if len(opt.img_size) == 1 else 1  # expand
20 |     print(opt)
21 | 
22 |     # Input
23 |     img = torch.zeros((opt.batch_size, 3, *opt.img_size))  # image size(1,3,320,192) iDetection
24 | 
25 |     # Load PyTorch model
26 |     attempt_download(opt.weights)
27 |     model = torch.load(opt.weights, map_location=torch.device('cpu'))['model'].float()
28 |     model.eval()
29 |     model.model[-1].export = True  # set Detect() layer export=True
30 |     y = model(img)  # dry run
31 | 
32 |     # TorchScript export
33 |     try:
34 |         print('\nStarting TorchScript export with torch %s...' % torch.__version__)
35 |         f = opt.weights.replace('.pt', '.torchscript.pt')  # filename
36 |         ts = torch.jit.trace(model, img)
37 |         ts.save(f)
38 |         print('TorchScript export success, saved as %s' % f)
39 |     except Exception as e:
40 |         print('TorchScript export failure: %s' % e)
41 | 
42 |     # ONNX export
43 |     try:
44 |         import onnx
45 | 
46 |         print('\nStarting ONNX export with onnx %s...' % onnx.__version__)
47 |         f = opt.weights.replace('.pt', '.onnx')  # filename
48 |         model.fuse()  # only for ONNX
49 |         torch.onnx.export(model, img, f, verbose=False, opset_version=12, input_names=['images'],
50 |                           output_names=['classes', 'boxes'] if y is None else ['output'])
51 | 
52 |         # Checks
53 |         onnx_model = onnx.load(f)  # load onnx model
54 |         onnx.checker.check_model(onnx_model)  # check onnx model
55 |         print(onnx.helper.printable_graph(onnx_model.graph))  # print a human readable model
56 |         print('ONNX export success, saved as %s' % f)
57 |     except Exception as e:
58 |         print('ONNX export failure: %s' % e)
59 | 
60 |     # CoreML export
61 |     try:
62 |         import coremltools as ct
63 | 
64 |         print('\nStarting CoreML export with coremltools %s...' % ct.__version__)
65 |         # convert model from torchscript and apply pixel scaling as per detect.py
66 |         model = ct.convert(ts, inputs=[ct.ImageType(name='images', shape=img.shape, scale=1 / 255.0, bias=[0, 0, 0])])
67 |         f = opt.weights.replace('.pt', '.mlmodel')  # filename
68 |         model.save(f)
69 |         print('CoreML export success, saved as %s' % f)
70 |     except Exception as e:
71 |         print('CoreML export failure: %s' % e)
72 | 
73 |     # Finish
74 |     print('\nExport complete. Visualize with https://github.com/lutzroeder/netron.')
75 | 


--------------------------------------------------------------------------------
/models/hub/yolov3-spp.yaml:
--------------------------------------------------------------------------------
 1 | # parameters
 2 | nc: 80  # number of classes
 3 | depth_multiple: 1.0  # model depth multiple
 4 | width_multiple: 1.0  # layer channel multiple
 5 | 
 6 | # anchors
 7 | anchors:
 8 |   - [10,13, 16,30, 33,23]  # P3/8
 9 |   - [30,61, 62,45, 59,119]  # P4/16
10 |   - [116,90, 156,198, 373,326]  # P5/32
11 | 
12 | # darknet53 backbone
13 | backbone:
14 |   # [from, number, module, args]
15 |   [[-1, 1, Conv, [32, 3, 1]],  # 0
16 |    [-1, 1, Conv, [64, 3, 2]],  # 1-P1/2
17 |    [-1, 1, Bottleneck, [64]],
18 |    [-1, 1, Conv, [128, 3, 2]],  # 3-P2/4
19 |    [-1, 2, Bottleneck, [128]],
20 |    [-1, 1, Conv, [256, 3, 2]],  # 5-P3/8
21 |    [-1, 8, Bottleneck, [256]],
22 |    [-1, 1, Conv, [512, 3, 2]],  # 7-P4/16
23 |    [-1, 8, Bottleneck, [512]],
24 |    [-1, 1, Conv, [1024, 3, 2]], # 9-P5/32
25 |    [-1, 4, Bottleneck, [1024]],  # 10
26 |   ]
27 | 
28 | # YOLOv3-SPP head
29 | head:
30 |   [[-1, 1, Bottleneck, [1024, False]],
31 |    [-1, 1, SPP, [512, [5, 9, 13]]],
32 |    [-1, 1, Conv, [1024, 3, 1]],
33 |    [-1, 1, Conv, [512, 1, 1]],
34 |    [-1, 1, Conv, [1024, 3, 1]],  # 15 (P5/32-large)
35 | 
36 |    [-2, 1, Conv, [256, 1, 1]],
37 |    [-1, 1, nn.Upsample, [None, 2, 'nearest']],
38 |    [[-1, 8], 1, Concat, [1]],  # cat backbone P4
39 |    [-1, 1, Bottleneck, [512, False]],
40 |    [-1, 1, Bottleneck, [512, False]],
41 |    [-1, 1, Conv, [256, 1, 1]],
42 |    [-1, 1, Conv, [512, 3, 1]],  # 22 (P4/16-medium)
43 | 
44 |    [-2, 1, Conv, [128, 1, 1]],
45 |    [-1, 1, nn.Upsample, [None, 2, 'nearest']],
46 |    [[-1, 6], 1, Concat, [1]],  # cat backbone P3
47 |    [-1, 1, Bottleneck, [256, False]],
48 |    [-1, 2, Bottleneck, [256, False]],  # 27 (P3/8-small)
49 | 
50 |    [[27, 22, 15], 1, Detect, [nc, anchors]],   # Detect(P3, P4, P5)
51 |   ]
52 | 


--------------------------------------------------------------------------------
/models/hub/yolov5-fpn.yaml:
--------------------------------------------------------------------------------
 1 | # parameters
 2 | nc: 80  # number of classes
 3 | depth_multiple: 1.0  # model depth multiple
 4 | width_multiple: 1.0  # layer channel multiple
 5 | 
 6 | # anchors
 7 | anchors:
 8 |   - [10,13, 16,30, 33,23]  # P3/8
 9 |   - [30,61, 62,45, 59,119]  # P4/16
10 |   - [116,90, 156,198, 373,326]  # P5/32
11 | 
12 | # YOLOv5 backbone
13 | backbone:
14 |   # [from, number, module, args]
15 |   [[-1, 1, Focus, [64, 3]],  # 0-P1/2
16 |    [-1, 1, Conv, [128, 3, 2]],  # 1-P2/4
17 |    [-1, 3, Bottleneck, [128]],
18 |    [-1, 1, Conv, [256, 3, 2]],  # 3-P3/8
19 |    [-1, 9, BottleneckCSP, [256]],
20 |    [-1, 1, Conv, [512, 3, 2]],  # 5-P4/16
21 |    [-1, 9, BottleneckCSP, [512]],
22 |    [-1, 1, Conv, [1024, 3, 2]], # 7-P5/32
23 |    [-1, 1, SPP, [1024, [5, 9, 13]]],
24 |    [-1, 6, BottleneckCSP, [1024]],  # 9
25 |   ]
26 | 
27 | # YOLOv5 FPN head
28 | head:
29 |   [[-1, 3, BottleneckCSP, [1024, False]],  # 10 (P5/32-large)
30 | 
31 |    [-1, 1, nn.Upsample, [None, 2, 'nearest']],
32 |    [[-1, 6], 1, Concat, [1]],  # cat backbone P4
33 |    [-1, 1, Conv, [512, 1, 1]],
34 |    [-1, 3, BottleneckCSP, [512, False]],  # 14 (P4/16-medium)
35 | 
36 |    [-1, 1, nn.Upsample, [None, 2, 'nearest']],
37 |    [[-1, 4], 1, Concat, [1]],  # cat backbone P3
38 |    [-1, 1, Conv, [256, 1, 1]],
39 |    [-1, 3, BottleneckCSP, [256, False]],  # 18 (P3/8-small)
40 | 
41 |    [[18, 14, 10], 1, Detect, [nc, anchors]],  # Detect(P3, P4, P5)
42 |   ]
43 | 


--------------------------------------------------------------------------------
/models/hub/yolov5-panet.yaml:
--------------------------------------------------------------------------------
 1 | # parameters
 2 | nc: 80  # number of classes
 3 | depth_multiple: 1.0  # model depth multiple
 4 | width_multiple: 1.0  # layer channel multiple
 5 | 
 6 | # anchors
 7 | anchors:
 8 |   - [116,90, 156,198, 373,326]  # P5/32
 9 |   - [30,61, 62,45, 59,119]  # P4/16
10 |   - [10,13, 16,30, 33,23]  # P3/8
11 | 
12 | # YOLOv5 backbone
13 | backbone:
14 |   # [from, number, module, args]
15 |   [[-1, 1, Focus, [64, 3]],  # 0-P1/2
16 |    [-1, 1, Conv, [128, 3, 2]],  # 1-P2/4
17 |    [-1, 3, BottleneckCSP, [128]],
18 |    [-1, 1, Conv, [256, 3, 2]],  # 3-P3/8
19 |    [-1, 9, BottleneckCSP, [256]],
20 |    [-1, 1, Conv, [512, 3, 2]],  # 5-P4/16
21 |    [-1, 9, BottleneckCSP, [512]],
22 |    [-1, 1, Conv, [1024, 3, 2]], # 7-P5/32
23 |    [-1, 1, SPP, [1024, [5, 9, 13]]],
24 |    [-1, 3, BottleneckCSP, [1024, False]],  # 9
25 |   ]
26 | 
27 | # YOLOv5 PANet head
28 | head:
29 |   [[-1, 1, Conv, [512, 1, 1]],
30 |    [-1, 1, nn.Upsample, [None, 2, 'nearest']],
31 |    [[-1, 6], 1, Concat, [1]],  # cat backbone P4
32 |    [-1, 3, BottleneckCSP, [512, False]],  # 13
33 | 
34 |    [-1, 1, Conv, [256, 1, 1]],
35 |    [-1, 1, nn.Upsample, [None, 2, 'nearest']],
36 |    [[-1, 4], 1, Concat, [1]],  # cat backbone P3
37 |    [-1, 3, BottleneckCSP, [256, False]],  # 17 (P3/8-small)
38 | 
39 |    [-1, 1, Conv, [256, 3, 2]],
40 |    [[-1, 14], 1, Concat, [1]],  # cat head P4
41 |    [-1, 3, BottleneckCSP, [512, False]],  # 20 (P4/16-medium)
42 | 
43 |    [-1, 1, Conv, [512, 3, 2]],
44 |    [[-1, 10], 1, Concat, [1]],  # cat head P5
45 |    [-1, 3, BottleneckCSP, [1024, False]],  # 23 (P5/32-large)
46 | 
47 |    [[17, 20, 23], 1, Detect, [nc, anchors]],  # Detect(P5, P4, P3)
48 |   ]
49 | 


--------------------------------------------------------------------------------
/models/yolo.py:
--------------------------------------------------------------------------------
  1 | import argparse
  2 | import math
  3 | from copy import deepcopy
  4 | from pathlib import Path
  5 | 
  6 | import torch
  7 | import torch.nn as nn
  8 | 
  9 | from models.common import Conv, Bottleneck, SPP, DWConv, Focus, BottleneckCSP, Concat,C3
 10 | from models.experimental import MixConv2d, CrossConv
 11 | from utils.general import check_anchor_order, make_divisible, check_file
 12 | from utils.torch_utils import (
 13 |     time_synchronized, fuse_conv_and_bn, model_info, initialize_weights, select_device)
 14 | 
 15 | 
 16 | class Detect(nn.Module):
 17 |     def __init__(self, nc=80, anchors=(), ch=()):  # detection layer
 18 |         super(Detect, self).__init__()
 19 |         self.stride = None  # strides computed during build
 20 |         self.nc = nc  # number of classes
 21 |         self.no = nc + 5  # number of outputs per anchor
 22 |         self.nl = len(anchors)  # number of detection layers
 23 |         self.na = len(anchors[0]) // 2  # number of anchors
 24 |         self.grid = [torch.zeros(1)] * self.nl  # init grid
 25 |         a = torch.tensor(anchors).float().view(self.nl, -1, 2)
 26 |         self.register_buffer('anchors', a)  # shape(nl,na,2)
 27 |         self.register_buffer('anchor_grid', a.clone().view(self.nl, 1, -1, 1, 1, 2))  # shape(nl,1,na,1,1,2)
 28 |         self.m = nn.ModuleList(nn.Conv2d(x, self.no * self.na, 1) for x in ch)  # output conv
 29 |         self.export = False  # onnx export
 30 | 
 31 |     def forward(self, x):
 32 |         # x = x.copy()  # for profiling
 33 |         z = []  # inference output
 34 |         self.training |= self.export
 35 |         for i in range(self.nl):
 36 |             x[i] = self.m[i](x[i])  # conv
 37 |             bs, _, ny, nx = x[i].shape  # x(bs,255,20,20) to x(bs,3,20,20,85)
 38 |             x[i] = x[i].view(bs, self.na, self.no, ny, nx).permute(0, 1, 3, 4, 2).contiguous()
 39 | 
 40 |             if not self.training:  # inference
 41 |                 if self.grid[i].shape[2:4] != x[i].shape[2:4]:
 42 |                     self.grid[i] = self._make_grid(nx, ny).to(x[i].device)
 43 | 
 44 |                 y = x[i].sigmoid()
 45 |                 y[..., 0:2] = (y[..., 0:2] * 2. - 0.5 + self.grid[i].to(x[i].device)) * self.stride[i]  # xy
 46 |                 y[..., 2:4] = (y[..., 2:4] * 2) ** 2 * self.anchor_grid[i]  # wh
 47 |                 z.append(y.view(bs, -1, self.no))
 48 | 
 49 |         return x if self.training else (torch.cat(z, 1), x)
 50 | 
 51 |     @staticmethod
 52 |     def _make_grid(nx=20, ny=20):
 53 |         yv, xv = torch.meshgrid([torch.arange(ny), torch.arange(nx)])
 54 |         return torch.stack((xv, yv), 2).view((1, 1, ny, nx, 2)).float()
 55 | 
 56 | 
 57 | class Model(nn.Module):
 58 |     def __init__(self, cfg='yolov5s.yaml', ch=3, nc=None):  # model, input channels, number of classes
 59 |         super(Model, self).__init__()
 60 |         if isinstance(cfg, dict):
 61 |             self.yaml = cfg  # model dict
 62 |         else:  # is *.yaml
 63 |             import yaml  # for torch hub
 64 |             self.yaml_file = Path(cfg).name
 65 |             with open(cfg) as f:
 66 |                 self.yaml = yaml.load(f, Loader=yaml.FullLoader)  # model dict
 67 | 
 68 |         # Define model
 69 |         if nc and nc != self.yaml['nc']:
 70 |             print('Overriding %s nc=%g with nc=%g' % (cfg, self.yaml['nc'], nc))
 71 |             self.yaml['nc'] = nc  # override yaml value
 72 |         self.model, self.save = parse_model(deepcopy(self.yaml), ch=[ch])  # model, savelist, ch_out
 73 |         # print([x.shape for x in self.forward(torch.zeros(1, ch, 64, 64))])
 74 | 
 75 |         # Build strides, anchors
 76 |         m = self.model[-1]  # Detect()
 77 |         if isinstance(m, Detect):
 78 |             s = 128  # 2x min stride
 79 |             m.stride = torch.tensor([s / x.shape[-2] for x in self.forward(torch.zeros(1, ch, s, s))])  # forward
 80 |             m.anchors /= m.stride.view(-1, 1, 1)
 81 |             check_anchor_order(m)
 82 |             self.stride = m.stride
 83 |             self._initialize_biases()  # only run once
 84 |             # print('Strides: %s' % m.stride.tolist())
 85 | 
 86 |         # Init weights, biases
 87 |         initialize_weights(self)
 88 |         self.info()
 89 |         print('')
 90 | 
 91 |     def forward(self, x, augment=False, profile=False):
 92 |         if augment:
 93 |             img_size = x.shape[-2:]  # height, width
 94 |             s = [1, 0.83, 0.67]  # scales
 95 |             f = [None, 3, None]  # flips (2-ud, 3-lr)
 96 |             y = []  # outputs
 97 |             for si, fi in zip(s, f):
 98 |                 xi = scale_img(x.flip(fi) if fi else x, si)
 99 |                 yi = self.forward_once(xi)[0]  # forward
100 |                 # cv2.imwrite('img%g.jpg' % s, 255 * xi[0].numpy().transpose((1, 2, 0))[:, :, ::-1])  # save
101 |                 yi[..., :4] /= si  # de-scale
102 |                 if fi == 2:
103 |                     yi[..., 1] = img_size[0] - yi[..., 1]  # de-flip ud
104 |                 elif fi == 3:
105 |                     yi[..., 0] = img_size[1] - yi[..., 0]  # de-flip lr
106 |                 y.append(yi)
107 |             return torch.cat(y, 1), None  # augmented inference, train
108 |         else:
109 |             return self.forward_once(x, profile)  # single-scale inference, train
110 | 
111 |     def forward_once(self, x, profile=False):
112 |         y, dt = [], []  # outputs
113 |         for m in self.model:
114 |             if m.f != -1:  # if not from previous layer
115 |                 x = y[m.f] if isinstance(m.f, int) else [x if j == -1 else y[j] for j in m.f]  # from earlier layers
116 | 
117 |             if profile:
118 |                 try:
119 |                     import thop
120 |                     o = thop.profile(m, inputs=(x,), verbose=False)[0] / 1E9 * 2  # FLOPS
121 |                 except:
122 |                     o = 0
123 |                 t = time_synchronized()
124 |                 for _ in range(10):
125 |                     _ = m(x)
126 |                 dt.append((time_synchronized() - t) * 100)
127 |                 print('%10.1f%10.0f%10.1fms %-40s' % (o, m.np, dt[-1], m.type))
128 | 
129 |             x = m(x)  # run
130 |             y.append(x if m.i in self.save else None)  # save output
131 | 
132 |         if profile:
133 |             print('%.1fms total' % sum(dt))
134 |         return x
135 | 
136 |     def _initialize_biases(self, cf=None):  # initialize biases into Detect(), cf is class frequency
137 |         # cf = torch.bincount(torch.tensor(np.concatenate(dataset.labels, 0)[:, 0]).long(), minlength=nc) + 1.
138 |         m = self.model[-1]  # Detect() module
139 |         for mi, s in zip(m.m, m.stride):  # from
140 |             b = mi.bias.view(m.na, -1)  # conv.bias(255) to (3,85)
141 |             b[:, 4] += math.log(8 / (640 / s) ** 2)  # obj (8 objects per 640 image)
142 |             b[:, 5:] += math.log(0.6 / (m.nc - 0.99)) if cf is None else torch.log(cf / cf.sum())  # cls
143 |             mi.bias = torch.nn.Parameter(b.view(-1), requires_grad=True)
144 | 
145 |     def _print_biases(self):
146 |         m = self.model[-1]  # Detect() module
147 |         for mi in m.m:  # from
148 |             b = mi.bias.detach().view(m.na, -1).T  # conv.bias(255) to (3,85)
149 |             print(('%6g Conv2d.bias:' + '%10.3g' * 6) % (mi.weight.shape[1], *b[:5].mean(1).tolist(), b[5:].mean()))
150 | 
151 |     # def _print_weights(self):
152 |     #     for m in self.model.modules():
153 |     #         if type(m) is Bottleneck:
154 |     #             print('%10.3g' % (m.w.detach().sigmoid() * 2))  # shortcut weights
155 | 
156 |     # def fuse(self):  # fuse model Conv2d() + BatchNorm2d() layers
157 |     #     print('Fusing layers... ', end='')
158 |     #     for m in self.model.modules():
159 |     #         if type(m) is Conv:
160 |     #             m._non_persistent_buffers_set = set()  # pytorch 1.6.0 compatability
161 |     #             m.conv = fuse_conv_and_bn(m.conv, m.bn)  # update conv
162 |     #             m.bn = None  # remove batchnorm
163 |     #             m.forward = m.fuseforward  # update forward
164 |     #     self.info()
165 |     #     return self
166 | 
167 |     def info(self):  # print model information
168 |         model_info(self)
169 | 
170 | 
171 | def parse_model(d, ch):  # model_dict, input_channels(3)
172 |     print('\n%3s%18s%3s%10s  %-40s%-30s' % ('', 'from', 'n', 'params', 'module', 'arguments'))
173 |     anchors, nc, gd, gw = d['anchors'], d['nc'], d['depth_multiple'], d['width_multiple']
174 |     na = (len(anchors[0]) // 2) if isinstance(anchors, list) else anchors  # number of anchors
175 |     no = na * (nc + 5)  # number of outputs = anchors * (classes + 5)
176 | 
177 |     layers, save, c2 = [], [], ch[-1]  # layers, savelist, ch out
178 |     for i, (f, n, m, args) in enumerate(d['backbone'] + d['head']):  # from, number, module, args
179 |         m = eval(m) if isinstance(m, str) else m  # eval strings
180 |         for j, a in enumerate(args):
181 |             try:
182 |                 args[j] = eval(a) if isinstance(a, str) else a  # eval strings
183 |             except:
184 |                 pass
185 | 
186 |         n = max(round(n * gd), 1) if n > 1 else n  # depth gain
187 |         if m in [nn.Conv2d, Conv, Bottleneck, SPP, DWConv, MixConv2d, Focus, CrossConv, BottleneckCSP, C3]:
188 |             c1, c2 = ch[f], args[0]
189 | 
190 |             # Normal
191 |             # if i > 0 and args[0] != no:  # channel expansion factor
192 |             #     ex = 1.75  # exponential (default 2.0)
193 |             #     e = math.log(c2 / ch[1]) / math.log(2)
194 |             #     c2 = int(ch[1] * ex ** e)
195 |             # if m != Focus:
196 | 
197 |             c2 = make_divisible(c2 * gw, 8) if c2 != no else c2
198 | 
199 |             # Experimental
200 |             # if i > 0 and args[0] != no:  # channel expansion factor
201 |             #     ex = 1 + gw  # exponential (default 2.0)
202 |             #     ch1 = 32  # ch[1]
203 |             #     e = math.log(c2 / ch1) / math.log(2)  # level 1-n
204 |             #     c2 = int(ch1 * ex ** e)
205 |             # if m != Focus:
206 |             #     c2 = make_divisible(c2, 8) if c2 != no else c2
207 | 
208 |             args = [c1, c2, *args[1:]]
209 |             if m in [BottleneckCSP, C3]:
210 |                 args.insert(2, n)
211 |                 n = 1
212 |         elif m is nn.BatchNorm2d:
213 |             args = [ch[f]]
214 |         elif m is Concat:
215 |             c2 = sum([ch[-1 if x == -1 else x + 1] for x in f])
216 |         elif m is Detect:
217 |             args.append([ch[x + 1] for x in f])
218 |             if isinstance(args[1], int):  # number of anchors
219 |                 args[1] = [list(range(args[1] * 2))] * len(f)
220 |         else:
221 |             c2 = ch[f]
222 | 
223 |         m_ = nn.Sequential(*[m(*args) for _ in range(n)]) if n > 1 else m(*args)  # module
224 |         t = str(m)[8:-2].replace('__main__.', '')  # module type
225 |         np = sum([x.numel() for x in m_.parameters()])  # number params
226 |         m_.i, m_.f, m_.type, m_.np = i, f, t, np  # attach index, 'from' index, type, number params
227 |         print('%3s%18s%3s%10.0f  %-40s%-30s' % (i, f, n, np, t, args))  # print
228 |         save.extend(x % i for x in ([f] if isinstance(f, int) else f) if x != -1)  # append to savelist
229 |         layers.append(m_)
230 |         ch.append(c2)
231 |     return nn.Sequential(*layers), sorted(save)
232 | 
233 | 
234 | if __name__ == '__main__':
235 |     parser = argparse.ArgumentParser()
236 |     parser.add_argument('--cfg', type=str, default='yolov5s.yaml', help='model.yaml')
237 |     parser.add_argument('--device', default='', help='cuda device, i.e. 0 or 0,1,2,3 or cpu')
238 |     opt = parser.parse_args()
239 |     opt.cfg = check_file(opt.cfg)  # check file
240 |     device = select_device(opt.device)
241 | 
242 |     # Create model
243 |     model = Model(opt.cfg).to(device)
244 |     model.train()
245 | 
246 |     # Profile
247 |     # img = torch.rand(8 if torch.cuda.is_available() else 1, 3, 640, 640).to(device)
248 |     # y = model(img, profile=True)
249 | 
250 |     # ONNX export
251 |     # model.model[-1].export = True
252 |     # torch.onnx.export(model, img, opt.cfg.replace('.yaml', '.onnx'), verbose=True, opset_version=11)
253 | 
254 |     # Tensorboard
255 |     # from torch.utils.tensorboard import SummaryWriter
256 |     # tb_writer = SummaryWriter()
257 |     # print("Run 'tensorboard --logdir=models/runs' to view tensorboard at http://localhost:6006/")
258 |     # tb_writer.add_graph(model.model, img)  # add model to tensorboard
259 |     # tb_writer.add_image('test', img[0], dataformats='CWH')  # add model to tensorboard
260 | 


--------------------------------------------------------------------------------
/models/yolov5s_v4.yaml:
--------------------------------------------------------------------------------
 1 | # parameters
 2 | nc: 80  # number of classes
 3 | depth_multiple: 0.33  # model depth multiple
 4 | width_multiple: 0.50  # layer channel multiple
 5 | 
 6 | # anchors
 7 | anchors:
 8 |   - [10,13, 16,30, 33,23]  # P3/8
 9 |   - [30,61, 62,45, 59,119]  # P4/16
10 |   - [116,90, 156,198, 373,326]  # P5/32
11 | 
12 | # YOLOv5 backbone
13 | backbone:
14 |   # [from, number, module, args]
15 |   [[-1, 1, Focus, [64, 3]],  # 0-P1/2
16 |    [-1, 1, Conv, [128, 3, 2]],  # 1-P2/4
17 |    [-1, 3, C3, [128]],
18 |    [-1, 1, Conv, [256, 3, 2]],  # 3-P3/8
19 |    [-1, 9, C3, [256]],
20 |    [-1, 1, Conv, [512, 3, 2]],  # 5-P4/16
21 |    [-1, 9, C3, [512]],
22 |    [-1, 1, Conv, [1024, 3, 2]],  # 7-P5/32
23 |    [-1, 1, SPP, [1024, [5, 9, 13]]],
24 |    [-1, 3, C3, [1024, False]],  # 9
25 |   ]
26 | 
27 | # YOLOv5 head
28 | head:
29 |   [[-1, 1, Conv, [512, 1, 1]],
30 |    [-1, 1, nn.Upsample, [None, 2, 'nearest']],
31 |    [[-1, 6], 1, Concat, [1]],  # cat backbone P4
32 |    [-1, 3, C3, [512, False]],  # 13
33 | 
34 |    [-1, 1, Conv, [256, 1, 1]],
35 |    [-1, 1, nn.Upsample, [None, 2, 'nearest']],
36 |    [[-1, 4], 1, Concat, [1]],  # cat backbone P3
37 |    [-1, 3, C3, [256, False]],  # 17 (P3/8-small)
38 | 
39 |    [-1, 1, Conv, [256, 3, 2]],
40 |    [[-1, 14], 1, Concat, [1]],  # cat head P4
41 |    [-1, 3, C3, [512, False]],  # 20 (P4/16-medium)
42 | 
43 |    [-1, 1, Conv, [512, 3, 2]],
44 |    [[-1, 10], 1, Concat, [1]],  # cat head P5
45 |    [-1, 3, C3, [1024, False]],  # 23 (P5/32-large)
46 | 
47 |    [[17, 20, 23], 1, Detect, [nc, anchors]],  # Detect(P3, P4, P5)
48 |   ]
49 | 


--------------------------------------------------------------------------------
/prune_yolov5s.sh:
--------------------------------------------------------------------------------
1 | python prune_yolov5s.py --cfg cfg/yolov5s_v4_hand.cfg --data data/oxfordhand.data --weights weights/last_v4s.pt --percent 0.8 --img_size 640


--------------------------------------------------------------------------------
/slim_prune_yolov5s_8x.sh:
--------------------------------------------------------------------------------
1 | python slim_prune_yolov5s_8x.py --cfg cfg/yolov5s_v4_hand.cfg --data data/oxfordhand.data --weights weights/last_v4s.pt --global_percent 0.5 --layer_keep 0.01 --img_size 640


--------------------------------------------------------------------------------
/test.py:
--------------------------------------------------------------------------------
  1 | import argparse
  2 | import json
  3 | 
  4 | from torch.utils.data import DataLoader
  5 | 
  6 | from modelsori import *
  7 | from utils.datasets import *
  8 | from utils.utils import *
  9 | import torchvision
 10 | 
 11 | def box_iouv5(box1, box2):
 12 |     # https://github.com/pytorch/vision/blob/master/torchvision/ops/boxes.py
 13 |     """
 14 |     Return intersection-over-union (Jaccard index) of boxes.
 15 |     Both sets of boxes are expected to be in (x1, y1, x2, y2) format.
 16 |     Arguments:
 17 |         box1 (Tensor[N, 4])
 18 |         box2 (Tensor[M, 4])
 19 |     Returns:
 20 |         iou (Tensor[N, M]): the NxM matrix containing the pairwise
 21 |             IoU values for every element in boxes1 and boxes2
 22 |     """
 23 | 
 24 |     def box_area(box):
 25 |         # box = 4xn
 26 |         return (box[2] - box[0]) * (box[3] - box[1])
 27 | 
 28 |     area1 = box_area(box1.T)
 29 |     area2 = box_area(box2.T)
 30 | 
 31 |     # inter(N,M) = (rb(N,M,2) - lt(N,M,2)).clamp(0).prod(2)
 32 |     inter = (torch.min(box1[:, None, 2:], box2[:, 2:]) - torch.max(box1[:, None, :2], box2[:, :2])).clamp(0).prod(2)
 33 |     return inter / (area1[:, None] + area2 - inter)  # iou = inter / (area1 + area2 - inter)
 34 | 
 35 | def non_max_suppressionv5(prediction, conf_thres=0.1, iou_thres=0.6, merge=False, classes=None, agnostic=False):
 36 |     """Performs Non-Maximum Suppression (NMS) on inference results
 37 | 
 38 |     Returns:
 39 |          detections with shape: nx6 (x1, y1, x2, y2, conf, cls)
 40 |     """
 41 |     if prediction.dtype is torch.float16:
 42 |         prediction = prediction.float()  # to FP32
 43 | 
 44 |     nc = prediction[0].shape[1] - 5  # number of classes
 45 |     xc = prediction[..., 4] > conf_thres  # candidates
 46 | 
 47 |     # Settings
 48 |     min_wh, max_wh = 2, 4096  # (pixels) minimum and maximum box width and height
 49 |     max_det = 300  # maximum number of detections per image
 50 |     time_limit = 10.0  # seconds to quit after
 51 |     redundant = True  # require redundant detections
 52 |     multi_label = nc > 1  # multiple labels per box (adds 0.5ms/img)
 53 | 
 54 |     t = time.time()
 55 |     output = [None] * prediction.shape[0]
 56 |     for xi, x in enumerate(prediction):  # image index, image inference
 57 |         # Apply constraints
 58 |         # x[((x[..., 2:4] < min_wh) | (x[..., 2:4] > max_wh)).any(1), 4] = 0  # width-height
 59 |         x = x[xc[xi]]  # confidence
 60 | 
 61 |         # If none remain process next image
 62 |         if not x.shape[0]:
 63 |             continue
 64 | 
 65 |         # Compute conf
 66 |         x[:, 5:] *= x[:, 4:5]  # conf = obj_conf * cls_conf
 67 | 
 68 |         # Box (center x, center y, width, height) to (x1, y1, x2, y2)
 69 |         box = xywh2xyxy(x[:, :4])
 70 | 
 71 |         # Detections matrix nx6 (xyxy, conf, cls)
 72 |         if multi_label:
 73 |             i, j = (x[:, 5:] > conf_thres).nonzero(as_tuple=False).T
 74 |             x = torch.cat((box[i], x[i, j + 5, None], j[:, None].float()), 1)
 75 |         else:  # best class only
 76 |             conf, j = x[:, 5:].max(1, keepdim=True)
 77 |             x = torch.cat((box, conf, j.float()), 1)[conf.view(-1) > conf_thres]
 78 | 
 79 |         # Filter by class
 80 |         if classes:
 81 |             x = x[(x[:, 5:6] == torch.tensor(classes, device=x.device)).any(1)]
 82 | 
 83 |         # Apply finite constraint
 84 |         # if not torch.isfinite(x).all():
 85 |         #     x = x[torch.isfinite(x).all(1)]
 86 | 
 87 |         # If none remain process next image
 88 |         n = x.shape[0]  # number of boxes
 89 |         if not n:
 90 |             continue
 91 | 
 92 |         # Sort by confidence
 93 |         # x = x[x[:, 4].argsort(descending=True)]
 94 | 
 95 |         # Batched NMS
 96 |         c = x[:, 5:6] * (0 if agnostic else max_wh)  # classes
 97 |         boxes, scores = x[:, :4] + c, x[:, 4]  # boxes (offset by class), scores
 98 |         i = torchvision.ops.boxes.nms(boxes, scores, iou_thres)
 99 |         if i.shape[0] > max_det:  # limit detections
100 |             i = i[:max_det]
101 |         if merge and (1 < n < 3E3):  # Merge NMS (boxes merged using weighted mean)
102 |             try:  # update boxes as boxes(i,4) = weights(i,n) * boxes(n,4)
103 |                 iou = box_iouv5(boxes[i], boxes) > iou_thres  # iou matrix
104 |                 weights = iou * scores[None]  # box weights
105 |                 x[i, :4] = torch.mm(weights, x[:, :4]).float() / weights.sum(1, keepdim=True)  # merged boxes
106 |                 if redundant:
107 |                     i = i[iou.sum(1) > 1]  # require redundancy
108 |             except:  # possible CUDA error https://github.com/ultralytics/yolov3/issues/1139
109 |                 print(x, i, x.shape, i.shape)
110 |                 pass
111 | 
112 |         output[xi] = x[i]
113 |         if (time.time() - t) > time_limit:
114 |             break  # time limit exceeded
115 | 
116 |     return output
117 | 
118 | def test(cfg,
119 |          data,
120 |          weights=None,
121 |          batch_size=16,
122 |          img_size=416,
123 |          iou_thres=0.5,
124 |          conf_thres=0.001,
125 |          nms_thres=0.5,
126 |          save_json=False,
127 |          model=None):
128 |     
129 |     # Initialize/load model and set device
130 |     if model is None:
131 |         device = torch_utils.select_device(opt.device)
132 |         verbose = True
133 | 
134 |         # Initialize model
135 |         model = Darknet(cfg, img_size).to(device)
136 | 
137 |         # Load weights
138 |         attempt_download(weights)
139 |         if weights.endswith('.pt'):  # pytorch format
140 |             model.load_state_dict(torch.load(weights, map_location=device)['model'])
141 |         else:  # darknet format
142 |             _ = load_darknet_weights(model, weights)
143 | 
144 |         if torch.cuda.device_count() > 1:
145 |             model = nn.DataParallel(model)
146 |     else:
147 |         device = next(model.parameters()).device  # get model device
148 |         verbose = False
149 | 
150 |     # Configure run
151 |     data = parse_data_cfg(data)
152 |     nc = int(data['classes'])  # number of classes
153 |     test_path = data['valid']  # path to test images
154 |     names = load_classes(data['names'])  # class names
155 | 
156 |     # Dataloader
157 |     dataset = LoadImagesAndLabels(test_path, img_size, batch_size)
158 |     dataloader = DataLoader(dataset,
159 |                             batch_size=batch_size,
160 |                             num_workers=min([os.cpu_count(), batch_size, 16]),
161 |                             pin_memory=True,
162 |                             collate_fn=dataset.collate_fn)
163 | 
164 |     seen = 0
165 |     model.eval()
166 |     coco91class = 3
167 |     s = ('%20s' + '%10s' * 6) % ('Class', 'Images', 'Targets', 'P', 'R', 'mAP', 'F1')
168 |     p, r, f1, mp, mr, map, mf1 = 0., 0., 0., 0., 0., 0., 0.
169 |     loss = torch.zeros(3)
170 |     jdict, stats, ap, ap_class = [], [], [], []
171 |     for batch_i, (imgs, targets, paths, shapes) in enumerate(tqdm(dataloader, desc=s)):
172 |         targets = targets.to(device)
173 |         imgs = imgs.to(device)
174 |         _, _, height, width = imgs.shape  # batch size, channels, height, width
175 | 
176 |         # Plot images with bounding boxes
177 |         if batch_i == 0 and not os.path.exists('test_batch0.jpg'):
178 |             plot_images(imgs=imgs, targets=targets, paths=paths, fname='test_batch0.jpg')
179 | 
180 |         # Run model
181 |         inf_out, train_out = model(imgs)  # inference and training outputs
182 | 
183 |         # Compute loss
184 |         if hasattr(model, 'hyp'):  # if model has loss hyperparameters
185 |             loss += compute_loss(train_out, targets, model)[1][:3].cpu()  # GIoU, obj, cls
186 | 
187 |         # Run NMS
188 |         # output = non_max_suppression(inf_out, conf_thres=conf_thres, nms_thres=nms_thres)
189 |         output = non_max_suppressionv5(inf_out,conf_thres=conf_thres, iou_thres=nms_thres, classes=None,agnostic=False)
190 | 
191 |         # Statistics per image
192 |         for si, pred in enumerate(output):
193 |             labels = targets[targets[:, 0] == si, 1:]
194 |             nl = len(labels)
195 |             tcls = labels[:, 0].tolist() if nl else []  # target class
196 |             seen += 1
197 | 
198 |             if pred is None:
199 |                 if nl:
200 |                     stats.append(([], torch.Tensor(), torch.Tensor(), tcls))
201 |                 continue
202 | 
203 |             # Append to text file
204 |             # with open('test.txt', 'a') as file:
205 |             #    [file.write('%11.5g' * 7 % tuple(x) + '\n') for x in pred]
206 | 
207 |             # Append to pycocotools JSON dictionary
208 |             if save_json:
209 |                 # [{"image_id": 42, "category_id": 18, "bbox": [258.15, 41.29, 348.26, 243.78], "score": 0.236}, ...
210 |                 image_id = int(Path(paths[si]).stem.split('_')[-1])
211 |                 box = pred[:, :4].clone()  # xyxy
212 |                 scale_coords(imgs[si].shape[1:], box, shapes[si])  # to original shape
213 |                 box = xyxy2xywh(box)  # xywh
214 |                 box[:, :2] -= box[:, 2:] / 2  # xy center to top-left corner
215 |                 for di, d in enumerate(pred):
216 |                     jdict.append({'image_id': image_id,
217 |                                   'category_id': coco91class[int(d[6])],
218 |                                   'bbox': [floatn(x, 3) for x in box[di]],
219 |                                   'score': floatn(d[4], 5)})
220 | 
221 |             # Clip boxes to image bounds
222 |             clip_coords(pred, (height, width))
223 | 
224 |             # Assign all predictions as incorrect
225 |             correct = [0] * len(pred)
226 |             if nl:
227 |                 detected = []
228 |                 tcls_tensor = labels[:, 0]
229 | 
230 |                 # target boxes
231 |                 tbox = xywh2xyxy(labels[:, 1:5])
232 |                 tbox[:, [0, 2]] *= width
233 |                 tbox[:, [1, 3]] *= height
234 | 
235 |                 # Search for correct predictions
236 |                 for i, (*pbox, pconf, pcls) in enumerate(pred):
237 |                 # for i, (*pbox, pconf, pcls_conf, pcls) in enumerate(pred):
238 | 
239 |                     # Break if all targets already located in image
240 |                     if len(detected) == nl:
241 |                         break
242 | 
243 |                     # Continue if predicted class not among image classes
244 |                     if pcls.item() not in tcls:
245 |                         continue
246 | 
247 |                     # Best iou, index between pred and targets
248 |                     m = (pcls == tcls_tensor).nonzero().view(-1)
249 |                     iou, bi = bbox_iou(pbox, tbox[m]).max(0)
250 | 
251 |                     # If iou > threshold and class is correct mark as correct
252 |                     if iou > iou_thres and m[bi] not in detected:  # and pcls == tcls[bi]:
253 |                         correct[i] = 1
254 |                         detected.append(m[bi])
255 | 
256 |             # Append statistics (correct, conf, pcls, tcls)
257 |             # stats.append((correct, pred[:, 4].cpu(), pred[:, 6].cpu(), tcls))
258 |             stats.append((correct, pred[:, 4].cpu(), pred[:, 5].cpu(), tcls))
259 | 
260 |     # Compute statistics
261 |     stats = [np.concatenate(x, 0) for x in list(zip(*stats))]  # to numpy
262 |     if len(stats):
263 |         p, r, ap, f1, ap_class = ap_per_class(*stats)
264 |         mp, mr, map, mf1 = p.mean(), r.mean(), ap.mean(), f1.mean()
265 |         nt = np.bincount(stats[3].astype(np.int64), minlength=nc)  # number of targets per class
266 |     else:
267 |         nt = torch.zeros(1)
268 | 
269 |     # Print results
270 |     pf = '%20s' + '%10.3g' * 6  # print format
271 |     print(pf % ('all', seen, nt.sum(), mp, mr, map, mf1))
272 | 
273 |     # Print results per class
274 |     if verbose and nc > 1 and len(stats):
275 |         for i, c in enumerate(ap_class):
276 |             print(pf % (names[c], seen, nt[c], p[i], r[i], ap[i], f1[i]))
277 | 
278 |     # Save JSON
279 |     if save_json and map and len(jdict):
280 |         try:
281 |             imgIds = [int(Path(x).stem.split('_')[-1]) for x in dataset.img_files]
282 |             with open('results.json', 'w') as file:
283 |                 json.dump(jdict, file)
284 | 
285 |             from pycocotools.coco import COCO
286 |             from pycocotools.cocoeval import COCOeval
287 | 
288 |             # https://github.com/cocodataset/cocoapi/blob/master/PythonAPI/pycocoEvalDemo.ipynb
289 |             cocoGt = COCO('../coco/annotations/instances_val2014.json')  # initialize COCO ground truth api
290 |             cocoDt = cocoGt.loadRes('results.json')  # initialize COCO pred api
291 | 
292 |             cocoEval = COCOeval(cocoGt, cocoDt, 'bbox')
293 |             cocoEval.params.imgIds = imgIds  # [:32]  # only evaluate these images
294 |             cocoEval.evaluate()
295 |             cocoEval.accumulate()
296 |             cocoEval.summarize()
297 |             map = cocoEval.stats[1]  # update mAP to pycocotools mAP
298 |         except:
299 |             print('WARNING: missing dependency pycocotools from requirements.txt. Can not compute official COCO mAP.')
300 | 
301 |     # Return results
302 |     maps = np.zeros(nc) + map
303 |     for i, c in enumerate(ap_class):
304 |         maps[c] = ap[i]
305 |     return (mp, mr, map, mf1, *(loss / len(dataloader)).tolist()), maps
306 | 
307 | 
308 | if __name__ == '__main__':
309 |     parser = argparse.ArgumentParser(prog='test.py')
310 |     parser.add_argument('--cfg', type=str, default='cfg/yolov5s.cfg', help='cfg file path')
311 |     parser.add_argument('--data', type=str, default='data/coco.data', help='coco.data file path')
312 |     parser.add_argument('--weights', type=str, default='weights/yolov3-spp.weights', help='path to weights file')
313 |     parser.add_argument('--batch-size', type=int, default=16, help='size of each image batch')
314 |     parser.add_argument('--img-size', type=int, default=416, help='inference size (pixels)')
315 |     parser.add_argument('--iou-thres', type=float, default=0.5, help='iou threshold required to qualify as detected')
316 |     parser.add_argument('--conf-thres', type=float, default=0.001, help='object confidence threshold')
317 |     parser.add_argument('--nms-thres', type=float, default=0.5, help='iou threshold for non-maximum suppression')
318 |     parser.add_argument('--save-json', action='store_true', help='save a cocoapi-compatible JSON results file')
319 |     parser.add_argument('--device', default='', help='device id (i.e. 0 or 0,1) or cpu')
320 |     opt = parser.parse_args()
321 |     print(opt)
322 | 
323 |     with torch.no_grad():
324 |         test(opt.cfg,
325 |              opt.data,
326 |              opt.weights,
327 |              opt.batch_size,
328 |              opt.img_size,
329 |              opt.iou_thres,
330 |              opt.conf_thres,
331 |              opt.nms_thres,
332 |              opt.save_json)
333 | 


--------------------------------------------------------------------------------
/test_yolov5s.py:
--------------------------------------------------------------------------------
  1 | from modelsori import *
  2 | from utils.utils import *
  3 | import numpy as np
  4 | from copy import deepcopy
  5 | from test import test
  6 | from terminaltables import AsciiTable
  7 | import time
  8 | from utils.prune_utils import *
  9 | import argparse
 10 | 
 11 | from models.yolo import Model
 12 | 
 13 | import torchvision
 14 | 
 15 | def letterboxv5(img, new_shape=(640, 640), color=(114, 114, 114), auto=True, scaleFill=False, scaleup=True):
 16 |     # Resize image to a 32-pixel-multiple rectangle https://github.com/ultralytics/yolov3/issues/232
 17 |     shape = img.shape[:2]  # current shape [height, width]
 18 |     if isinstance(new_shape, int):
 19 |         new_shape = (new_shape, new_shape)
 20 | 
 21 |     # Scale ratio (new / old)
 22 |     r = min(new_shape[0] / shape[0], new_shape[1] / shape[1])
 23 |     if not scaleup:  # only scale down, do not scale up (for better test mAP)
 24 |         r = min(r, 1.0)
 25 | 
 26 |     # Compute padding
 27 |     ratio = r, r  # width, height ratios
 28 |     new_unpad = int(round(shape[1] * r)), int(round(shape[0] * r))
 29 |     dw, dh = new_shape[1] - new_unpad[0], new_shape[0] - new_unpad[1]  # wh padding
 30 |     if auto:  # minimum rectangle
 31 |         # dw, dh = np.mod(dw, 64), np.mod(dh, 64)  # wh padding
 32 |         dw, dh = np.mod(dw, 32), np.mod(dh, 32)  # wh padding
 33 |     elif scaleFill:  # stretch
 34 |         dw, dh = 0.0, 0.0
 35 |         new_unpad = (new_shape[1], new_shape[0])
 36 |         ratio = new_shape[1] / shape[1], new_shape[0] / shape[0]  # width, height ratios
 37 | 
 38 |     dw /= 2  # divide padding into 2 sides
 39 |     dh /= 2
 40 | 
 41 |     if shape[::-1] != new_unpad:  # resize
 42 |         img = cv2.resize(img, new_unpad, interpolation=cv2.INTER_LINEAR)
 43 |     top, bottom = int(round(dh - 0.1)), int(round(dh + 0.1))
 44 |     left, right = int(round(dw - 0.1)), int(round(dw + 0.1))
 45 |     img = cv2.copyMakeBorder(img, top, bottom, left, right, cv2.BORDER_CONSTANT, value=color)  # add border
 46 |     return img, ratio, (dw, dh)
 47 | 
 48 | def box_iou(box1, box2):
 49 |     # https://github.com/pytorch/vision/blob/master/torchvision/ops/boxes.py
 50 |     """
 51 |     Return intersection-over-union (Jaccard index) of boxes.
 52 |     Both sets of boxes are expected to be in (x1, y1, x2, y2) format.
 53 |     Arguments:
 54 |         box1 (Tensor[N, 4])
 55 |         box2 (Tensor[M, 4])
 56 |     Returns:
 57 |         iou (Tensor[N, M]): the NxM matrix containing the pairwise
 58 |             IoU values for every element in boxes1 and boxes2
 59 |     """
 60 | 
 61 |     def box_area(box):
 62 |         # box = 4xn
 63 |         return (box[2] - box[0]) * (box[3] - box[1])
 64 | 
 65 |     area1 = box_area(box1.T)
 66 |     area2 = box_area(box2.T)
 67 | 
 68 |     # inter(N,M) = (rb(N,M,2) - lt(N,M,2)).clamp(0).prod(2)
 69 |     inter = (torch.min(box1[:, None, 2:], box2[:, 2:]) - torch.max(box1[:, None, :2], box2[:, :2])).clamp(0).prod(2)
 70 |     return inter / (area1[:, None] + area2 - inter)  # iou = inter / (area1 + area2 - inter)
 71 | 
 72 | def non_max_suppressionv5(prediction, conf_thres=0.1, iou_thres=0.6, merge=False, classes=None, agnostic=False):
 73 |     """Performs Non-Maximum Suppression (NMS) on inference results
 74 | 
 75 |     Returns:
 76 |          detections with shape: nx6 (x1, y1, x2, y2, conf, cls)
 77 |     """
 78 |     if prediction.dtype is torch.float16:
 79 |         prediction = prediction.float()  # to FP32
 80 | 
 81 |     nc = prediction[0].shape[1] - 5  # number of classes
 82 |     xc = prediction[..., 4] > conf_thres  # candidates
 83 | 
 84 |     # Settings
 85 |     min_wh, max_wh = 2, 4096  # (pixels) minimum and maximum box width and height
 86 |     max_det = 300  # maximum number of detections per image
 87 |     time_limit = 10.0  # seconds to quit after
 88 |     redundant = True  # require redundant detections
 89 |     multi_label = nc > 1  # multiple labels per box (adds 0.5ms/img)
 90 | 
 91 |     t = time.time()
 92 |     output = [None] * prediction.shape[0]
 93 |     for xi, x in enumerate(prediction):  # image index, image inference
 94 |         # Apply constraints
 95 |         # x[((x[..., 2:4] < min_wh) | (x[..., 2:4] > max_wh)).any(1), 4] = 0  # width-height
 96 |         x = x[xc[xi]]  # confidence
 97 | 
 98 |         # If none remain process next image
 99 |         if not x.shape[0]:
100 |             continue
101 | 
102 |         # Compute conf
103 |         x[:, 5:] *= x[:, 4:5]  # conf = obj_conf * cls_conf
104 | 
105 |         # Box (center x, center y, width, height) to (x1, y1, x2, y2)
106 |         box = xywh2xyxy(x[:, :4])
107 | 
108 |         # Detections matrix nx6 (xyxy, conf, cls)
109 |         if multi_label:
110 |             i, j = (x[:, 5:] > conf_thres).nonzero(as_tuple=False).T
111 |             x = torch.cat((box[i], x[i, j + 5, None], j[:, None].float()), 1)
112 |         else:  # best class only
113 |             conf, j = x[:, 5:].max(1, keepdim=True)
114 |             x = torch.cat((box, conf, j.float()), 1)[conf.view(-1) > conf_thres]
115 | 
116 |         # Filter by class
117 |         if classes:
118 |             x = x[(x[:, 5:6] == torch.tensor(classes, device=x.device)).any(1)]
119 | 
120 |         # Apply finite constraint
121 |         # if not torch.isfinite(x).all():
122 |         #     x = x[torch.isfinite(x).all(1)]
123 | 
124 |         # If none remain process next image
125 |         n = x.shape[0]  # number of boxes
126 |         if not n:
127 |             continue
128 | 
129 |         # Sort by confidence
130 |         # x = x[x[:, 4].argsort(descending=True)]
131 | 
132 |         # Batched NMS
133 |         c = x[:, 5:6] * (0 if agnostic else max_wh)  # classes
134 |         boxes, scores = x[:, :4] + c, x[:, 4]  # boxes (offset by class), scores
135 |         i = torchvision.ops.boxes.nms(boxes, scores, iou_thres)
136 |         if i.shape[0] > max_det:  # limit detections
137 |             i = i[:max_det]
138 |         if merge and (1 < n < 3E3):  # Merge NMS (boxes merged using weighted mean)
139 |             try:  # update boxes as boxes(i,4) = weights(i,n) * boxes(n,4)
140 |                 iou = box_iou(boxes[i], boxes) > iou_thres  # iou matrix
141 |                 weights = iou * scores[None]  # box weights
142 |                 x[i, :4] = torch.mm(weights, x[:, :4]).float() / weights.sum(1, keepdim=True)  # merged boxes
143 |                 if redundant:
144 |                     i = i[iou.sum(1) > 1]  # require redundancy
145 |             except:  # possible CUDA error https://github.com/ultralytics/yolov3/issues/1139
146 |                 print(x, i, x.shape, i.shape)
147 |                 pass
148 | 
149 |         output[xi] = x[i]
150 |         if (time.time() - t) > time_limit:
151 |             break  # time limit exceeded
152 | 
153 |     return output
154 | 
155 | def plot_one_box(x, img, color=None, label=None, line_thickness=None):
156 |     # Plots one bounding box on image img
157 |     tl = line_thickness or round(0.002 * (img.shape[0] + img.shape[1]) / 2) + 1  # line/font thickness
158 |     color = color or [random.randint(0, 255) for _ in range(3)]
159 |     c1, c2 = (int(x[0]), int(x[1])), (int(x[2]), int(x[3]))
160 |     cv2.rectangle(img, c1, c2, color, thickness=tl, lineType=cv2.LINE_AA)
161 |     if label:
162 |         tf = max(tl - 1, 1)  # font thickness
163 |         t_size = cv2.getTextSize(label, 0, fontScale=tl / 3, thickness=tf)[0]
164 |         c2 = c1[0] + t_size[0], c1[1] - t_size[1] - 3
165 |         cv2.rectangle(img, c1, c2, color, -1, cv2.LINE_AA)  # filled
166 |         cv2.putText(img, label, (c1[0], c1[1] - 2), 0, tl / 3, [225, 255, 255], thickness=tf, lineType=cv2.LINE_AA)
167 | 
168 | def copy_conv(conv_src,conv_dst):
169 |     conv_dst[0] = conv_src.conv
170 |     conv_dst[1] = conv_src.bn
171 |     conv_dst[2] = conv_src.act
172 | 
173 | def copy_weight_v4(modelyolov5,model):
174 |     focus = list(modelyolov5.model.children())[0]
175 |     copy_conv(focus.conv, model.module_list[1])
176 |     conv1 = list(modelyolov5.model.children())[1]
177 |     copy_conv(conv1, model.module_list[2])
178 |     cspnet1 = list(modelyolov5.model.children())[2]
179 |     copy_conv(cspnet1.cv2, model.module_list[3])
180 |     copy_conv(cspnet1.cv1, model.module_list[5])
181 |     copy_conv(cspnet1.m[0].cv1, model.module_list[6])
182 |     copy_conv(cspnet1.m[0].cv2, model.module_list[7])
183 |     copy_conv(cspnet1.cv3, model.module_list[10])
184 |     conv2 = list(modelyolov5.model.children())[3]
185 |     copy_conv(conv2, model.module_list[11])
186 |     cspnet2 = list(modelyolov5.model.children())[4]
187 |     copy_conv(cspnet2.cv2, model.module_list[12])
188 |     copy_conv(cspnet2.cv1, model.module_list[14])
189 |     copy_conv(cspnet2.m[0].cv1, model.module_list[15])
190 |     copy_conv(cspnet2.m[0].cv2, model.module_list[16])
191 |     copy_conv(cspnet2.m[1].cv1, model.module_list[18])
192 |     copy_conv(cspnet2.m[1].cv2, model.module_list[19])
193 |     copy_conv(cspnet2.m[2].cv1, model.module_list[21])
194 |     copy_conv(cspnet2.m[2].cv2, model.module_list[22])
195 |     copy_conv(cspnet2.cv3, model.module_list[25])
196 |     conv3 = list(modelyolov5.model.children())[5]
197 |     copy_conv(conv3, model.module_list[26])
198 |     cspnet3 = list(modelyolov5.model.children())[6]
199 |     copy_conv(cspnet3.cv2, model.module_list[27])
200 |     copy_conv(cspnet3.cv1, model.module_list[29])
201 |     copy_conv(cspnet3.m[0].cv1, model.module_list[30])
202 |     copy_conv(cspnet3.m[0].cv2, model.module_list[31])
203 |     copy_conv(cspnet3.m[1].cv1, model.module_list[33])
204 |     copy_conv(cspnet3.m[1].cv2, model.module_list[34])
205 |     copy_conv(cspnet3.m[2].cv1, model.module_list[36])
206 |     copy_conv(cspnet3.m[2].cv2, model.module_list[37])
207 |     copy_conv(cspnet3.cv3, model.module_list[40])
208 |     conv4 = list(modelyolov5.model.children())[7]
209 |     copy_conv(conv4, model.module_list[41])
210 |     spp = list(modelyolov5.model.children())[8]
211 |     copy_conv(spp.cv1, model.module_list[42])
212 |     model.module_list[43] = spp.m[0]
213 |     model.module_list[45] = spp.m[1]
214 |     model.module_list[47] = spp.m[2]
215 |     copy_conv(spp.cv2, model.module_list[49])
216 |     cspnet4 = list(modelyolov5.model.children())[9]
217 |     copy_conv(cspnet4.cv2, model.module_list[50])
218 |     copy_conv(cspnet4.cv1, model.module_list[52])
219 |     copy_conv(cspnet4.m[0].cv1, model.module_list[53])
220 |     copy_conv(cspnet4.m[0].cv2, model.module_list[54])
221 |     copy_conv(cspnet4.cv3, model.module_list[56])
222 |     conv5 = list(modelyolov5.model.children())[10]
223 |     copy_conv(conv5, model.module_list[57])
224 |     upsample1 = list(modelyolov5.model.children())[11]
225 |     model.module_list[58] = upsample1
226 |     cspnet5 = list(modelyolov5.model.children())[13]
227 |     copy_conv(cspnet5.cv2, model.module_list[60])
228 |     copy_conv(cspnet5.cv1, model.module_list[62])
229 |     copy_conv(cspnet5.m[0].cv1, model.module_list[63])
230 |     copy_conv(cspnet5.m[0].cv2, model.module_list[64])
231 |     copy_conv(cspnet5.cv3, model.module_list[66])
232 |     conv6 = list(modelyolov5.model.children())[14]
233 |     copy_conv(conv6, model.module_list[67])
234 |     upsample2 = list(modelyolov5.model.children())[15]
235 |     model.module_list[68] = upsample2
236 |     cspnet6 = list(modelyolov5.model.children())[17]
237 |     copy_conv(cspnet6.cv2, model.module_list[70])
238 |     copy_conv(cspnet6.cv1, model.module_list[72])
239 |     copy_conv(cspnet6.m[0].cv1, model.module_list[73])
240 |     copy_conv(cspnet6.m[0].cv2, model.module_list[74])
241 |     copy_conv(cspnet6.cv3, model.module_list[76])
242 |     conv7 = list(modelyolov5.model.children())[18]
243 |     copy_conv(conv7, model.module_list[80])
244 |     cspnet7 = list(modelyolov5.model.children())[20]
245 |     copy_conv(cspnet7.cv2, model.module_list[82])
246 |     copy_conv(cspnet7.cv1, model.module_list[84])
247 |     copy_conv(cspnet7.m[0].cv1, model.module_list[85])
248 |     copy_conv(cspnet7.m[0].cv2, model.module_list[86])
249 |     copy_conv(cspnet7.cv3, model.module_list[88])
250 |     conv8 = list(modelyolov5.model.children())[21]
251 |     copy_conv(conv8, model.module_list[92])
252 |     cspnet8 = list(modelyolov5.model.children())[23]
253 |     copy_conv(cspnet8.cv2, model.module_list[94])
254 |     copy_conv(cspnet8.cv1, model.module_list[96])
255 |     copy_conv(cspnet8.m[0].cv1, model.module_list[97])
256 |     copy_conv(cspnet8.m[0].cv2, model.module_list[98])
257 |     copy_conv(cspnet8.cv3, model.module_list[100])
258 |     detect = list(modelyolov5.model.children())[24]
259 |     model.module_list[77][0] = detect.m[0]
260 |     model.module_list[89][0] = detect.m[1]
261 |     model.module_list[101][0] = detect.m[2]
262 | 
263 | def initialize_weights(model):
264 |     for m in model.modules():
265 |         t = type(m)
266 |         if t is nn.Conv2d:
267 |             pass  # nn.init.kaiming_normal_(m.weight, mode='fan_out', nonlinearity='relu')
268 |         elif t is nn.BatchNorm2d:
269 |             m.eps = 1e-3
270 |             m.momentum = 0.03
271 |         elif t in [nn.LeakyReLU, nn.ReLU, nn.ReLU6]:
272 |             m.inplace = True
273 | 
274 | if __name__ == '__main__':
275 |     parser = argparse.ArgumentParser()
276 |     parser.add_argument('--cfg', type=str, default='cfg/yolov5s_v4.cfg', help='cfg file path')
277 |     parser.add_argument('--data', type=str, default='data/coco.data', help='*.data file path')
278 |     parser.add_argument('--weights', type=str, default='weights/yolov5s_v4.pt', help='sparse model weights')
279 |     parser.add_argument('--img_size', type=int, default=416, help='inference size (pixels)')
280 |     opt = parser.parse_args()
281 |     print(opt)
282 | 
283 |     img_size = opt.img_size
284 |     device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
285 | 
286 |     #the way of loading yolov5s
287 |     # ckpt = torch.load(opt.weights, map_location=device)  # load checkpoint
288 |     # modelyolov5 = Model('models/yolov5s_v4.yaml', nc=80).to(device)
289 |     # exclude = ['anchor']  # exclude keys
290 |     # ckpt['model'] = {k: v for k, v in ckpt['model'].float().state_dict().items()
291 |     #                  if k in modelyolov5.state_dict() and not any(x in k for x in exclude)
292 |     #                  and modelyolov5.state_dict()[k].shape == v.shape}
293 |     # modelyolov5.load_state_dict(ckpt['model'], strict=False)
294 | 
295 |     #another way of loading yolov5s
296 |     modelyolov5=torch.load(opt.weights, map_location=device)['model'].float().eval()
297 |     modelyolov5.model[24].export = False  # onnx export
298 | 
299 |     # model=modelyolov5
300 | 
301 |     #load yolov5s from cfg
302 |     model = Darknet(opt.cfg, (img_size, img_size)).to(device)
303 |     copy_weight_v4(modelyolov5,model)
304 | 
305 |     path='data/samples/bus.jpg'
306 |     img0 = cv2.imread(path)  # BGR
307 |     # Padded resize
308 |     img = letterboxv5(img0, new_shape=416)[0]
309 | 
310 |     # Convert
311 |     img = img[:, :, ::-1].transpose(2, 0, 1)  # BGR to RGB, to 3x416x416
312 |     img = np.ascontiguousarray(img)
313 |     img = torch.from_numpy(img).to(device)
314 |     img = img.float()
315 |     img /= 255.0  # 0 - 255 to 0.0 - 1.0
316 |     if img.ndimension() == 3:
317 |         img = img.unsqueeze(0)
318 | 
319 |     # modelyolov5.eval()
320 | 
321 | 
322 |     model.eval()
323 |     pred = model(img)[0]
324 | 
325 |     pred = non_max_suppressionv5(pred, 0.4, 0.5, classes=None,
326 |                                agnostic=False)
327 |     # Process detections
328 |     for i, det in enumerate(pred):  # detections per image
329 |         if det is not None and len(det):
330 |             # Rescale boxes from img_size to im0 size
331 |             det[:, :4] = scale_coords(img.shape[2:], det[:, :4], img0.shape).round()
332 | 
333 |             # Write results
334 |             for *xyxy, conf, cls in det:
335 |                 label = '%s %.2f' % (str(int(cls)), conf)
336 |                 plot_one_box(xyxy, img0, label=label, color=[random.randint(0, 255) for _ in range(3)], line_thickness=3)
337 |             cv2.imwrite("v5_cfg.jpg", img0)
338 | 
339 |     modelyolov5.eval()
340 |     pred = modelyolov5(img)[0]
341 | 
342 |     pred = non_max_suppressionv5(pred, 0.4, 0.5, classes=None,
343 |                                  agnostic=False)
344 |     # Process detections
345 |     for i, det in enumerate(pred):  # detections per image
346 |         if det is not None and len(det):
347 |             # Rescale boxes from img_size to im0 size
348 |             det[:, :4] = scale_coords(img.shape[2:], det[:, :4], img0.shape).round()
349 | 
350 |             # Write results
351 |             for *xyxy, conf, cls in det:
352 |                 label = '%s %.2f' % (str(int(cls)), conf)
353 |                 plot_one_box(xyxy, img0, label=label, color=[random.randint(0, 255) for _ in range(3)],
354 |                              line_thickness=3)
355 |             cv2.imwrite("v5.jpg", img0)
356 | 


--------------------------------------------------------------------------------
/tk1_time.xls:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/BokyLiu/YoloV5sl_V4_prune/c0ff39c5a5b10cbca95beb597c722cdc02e81885/tk1_time.xls


--------------------------------------------------------------------------------
/utils/__init__.py:
--------------------------------------------------------------------------------
1 | 
2 | 
3 | #


--------------------------------------------------------------------------------
/utils/adabound.py:
--------------------------------------------------------------------------------
  1 | import math
  2 | 
  3 | import torch
  4 | from torch.optim import Optimizer
  5 | 
  6 | 
  7 | class AdaBound(Optimizer):
  8 |     """Implements AdaBound algorithm.
  9 |     It has been proposed in `Adaptive Gradient Methods with Dynamic Bound of Learning Rate`_.
 10 |     Arguments:
 11 |         params (iterable): iterable of parameters to optimize or dicts defining
 12 |             parameter groups
 13 |         lr (float, optional): Adam learning rate (default: 1e-3)
 14 |         betas (Tuple[float, float], optional): coefficients used for computing
 15 |             running averages of gradient and its square (default: (0.9, 0.999))
 16 |         final_lr (float, optional): final (SGD) learning rate (default: 0.1)
 17 |         gamma (float, optional): convergence speed of the bound functions (default: 1e-3)
 18 |         eps (float, optional): term added to the denominator to improve
 19 |             numerical stability (default: 1e-8)
 20 |         weight_decay (float, optional): weight decay (L2 penalty) (default: 0)
 21 |         amsbound (boolean, optional): whether to use the AMSBound variant of this algorithm
 22 |     .. Adaptive Gradient Methods with Dynamic Bound of Learning Rate:
 23 |         https://openreview.net/forum?id=Bkg3g2R9FX
 24 |     """
 25 | 
 26 |     def __init__(self, params, lr=1e-3, betas=(0.9, 0.999), final_lr=0.1, gamma=1e-3,
 27 |                  eps=1e-8, weight_decay=0, amsbound=False):
 28 |         if not 0.0 <= lr:
 29 |             raise ValueError("Invalid learning rate: {}".format(lr))
 30 |         if not 0.0 <= eps:
 31 |             raise ValueError("Invalid epsilon value: {}".format(eps))
 32 |         if not 0.0 <= betas[0] < 1.0:
 33 |             raise ValueError("Invalid beta parameter at index 0: {}".format(betas[0]))
 34 |         if not 0.0 <= betas[1] < 1.0:
 35 |             raise ValueError("Invalid beta parameter at index 1: {}".format(betas[1]))
 36 |         if not 0.0 <= final_lr:
 37 |             raise ValueError("Invalid final learning rate: {}".format(final_lr))
 38 |         if not 0.0 <= gamma < 1.0:
 39 |             raise ValueError("Invalid gamma parameter: {}".format(gamma))
 40 |         defaults = dict(lr=lr, betas=betas, final_lr=final_lr, gamma=gamma, eps=eps,
 41 |                         weight_decay=weight_decay, amsbound=amsbound)
 42 |         super(AdaBound, self).__init__(params, defaults)
 43 | 
 44 |         self.base_lrs = list(map(lambda group: group['lr'], self.param_groups))
 45 | 
 46 |     def __setstate__(self, state):
 47 |         super(AdaBound, self).__setstate__(state)
 48 |         for group in self.param_groups:
 49 |             group.setdefault('amsbound', False)
 50 | 
 51 |     def step(self, closure=None):
 52 |         """Performs a single optimization step.
 53 |         Arguments:
 54 |             closure (callable, optional): A closure that reevaluates the model
 55 |                 and returns the loss.
 56 |         """
 57 |         loss = None
 58 |         if closure is not None:
 59 |             loss = closure()
 60 | 
 61 |         for group, base_lr in zip(self.param_groups, self.base_lrs):
 62 |             for p in group['params']:
 63 |                 if p.grad is None:
 64 |                     continue
 65 |                 grad = p.grad.data
 66 |                 if grad.is_sparse:
 67 |                     raise RuntimeError(
 68 |                         'Adam does not support sparse gradients, please consider SparseAdam instead')
 69 |                 amsbound = group['amsbound']
 70 | 
 71 |                 state = self.state[p]
 72 | 
 73 |                 # State initialization
 74 |                 if len(state) == 0:
 75 |                     state['step'] = 0
 76 |                     # Exponential moving average of gradient values
 77 |                     state['exp_avg'] = torch.zeros_like(p.data)
 78 |                     # Exponential moving average of squared gradient values
 79 |                     state['exp_avg_sq'] = torch.zeros_like(p.data)
 80 |                     if amsbound:
 81 |                         # Maintains max of all exp. moving avg. of sq. grad. values
 82 |                         state['max_exp_avg_sq'] = torch.zeros_like(p.data)
 83 | 
 84 |                 exp_avg, exp_avg_sq = state['exp_avg'], state['exp_avg_sq']
 85 |                 if amsbound:
 86 |                     max_exp_avg_sq = state['max_exp_avg_sq']
 87 |                 beta1, beta2 = group['betas']
 88 | 
 89 |                 state['step'] += 1
 90 | 
 91 |                 if group['weight_decay'] != 0:
 92 |                     grad = grad.add(group['weight_decay'], p.data)
 93 | 
 94 |                 # Decay the first and second moment running average coefficient
 95 |                 exp_avg.mul_(beta1).add_(1 - beta1, grad)
 96 |                 exp_avg_sq.mul_(beta2).addcmul_(1 - beta2, grad, grad)
 97 |                 if amsbound:
 98 |                     # Maintains the maximum of all 2nd moment running avg. till now
 99 |                     torch.max(max_exp_avg_sq, exp_avg_sq, out=max_exp_avg_sq)
100 |                     # Use the max. for normalizing running avg. of gradient
101 |                     denom = max_exp_avg_sq.sqrt().add_(group['eps'])
102 |                 else:
103 |                     denom = exp_avg_sq.sqrt().add_(group['eps'])
104 | 
105 |                 bias_correction1 = 1 - beta1 ** state['step']
106 |                 bias_correction2 = 1 - beta2 ** state['step']
107 |                 step_size = group['lr'] * math.sqrt(bias_correction2) / bias_correction1
108 | 
109 |                 # Applies bounds on actual learning rate
110 |                 # lr_scheduler cannot affect final_lr, this is a workaround to apply lr decay
111 |                 final_lr = group['final_lr'] * group['lr'] / base_lr
112 |                 lower_bound = final_lr * (1 - 1 / (group['gamma'] * state['step'] + 1))
113 |                 upper_bound = final_lr * (1 + 1 / (group['gamma'] * state['step']))
114 |                 step_size = torch.full_like(denom, step_size)
115 |                 step_size.div_(denom).clamp_(lower_bound, upper_bound).mul_(exp_avg)
116 | 
117 |                 p.data.add_(-step_size)
118 | 
119 |         return loss
120 | 
121 | 
122 | class AdaBoundW(Optimizer):
123 |     """Implements AdaBound algorithm with Decoupled Weight Decay (arxiv.org/abs/1711.05101)
124 |     It has been proposed in `Adaptive Gradient Methods with Dynamic Bound of Learning Rate`_.
125 |     Arguments:
126 |         params (iterable): iterable of parameters to optimize or dicts defining
127 |             parameter groups
128 |         lr (float, optional): Adam learning rate (default: 1e-3)
129 |         betas (Tuple[float, float], optional): coefficients used for computing
130 |             running averages of gradient and its square (default: (0.9, 0.999))
131 |         final_lr (float, optional): final (SGD) learning rate (default: 0.1)
132 |         gamma (float, optional): convergence speed of the bound functions (default: 1e-3)
133 |         eps (float, optional): term added to the denominator to improve
134 |             numerical stability (default: 1e-8)
135 |         weight_decay (float, optional): weight decay (L2 penalty) (default: 0)
136 |         amsbound (boolean, optional): whether to use the AMSBound variant of this algorithm
137 |     .. Adaptive Gradient Methods with Dynamic Bound of Learning Rate:
138 |         https://openreview.net/forum?id=Bkg3g2R9FX
139 |     """
140 | 
141 |     def __init__(self, params, lr=1e-3, betas=(0.9, 0.999), final_lr=0.1, gamma=1e-3,
142 |                  eps=1e-8, weight_decay=0, amsbound=False):
143 |         if not 0.0 <= lr:
144 |             raise ValueError("Invalid learning rate: {}".format(lr))
145 |         if not 0.0 <= eps:
146 |             raise ValueError("Invalid epsilon value: {}".format(eps))
147 |         if not 0.0 <= betas[0] < 1.0:
148 |             raise ValueError("Invalid beta parameter at index 0: {}".format(betas[0]))
149 |         if not 0.0 <= betas[1] < 1.0:
150 |             raise ValueError("Invalid beta parameter at index 1: {}".format(betas[1]))
151 |         if not 0.0 <= final_lr:
152 |             raise ValueError("Invalid final learning rate: {}".format(final_lr))
153 |         if not 0.0 <= gamma < 1.0:
154 |             raise ValueError("Invalid gamma parameter: {}".format(gamma))
155 |         defaults = dict(lr=lr, betas=betas, final_lr=final_lr, gamma=gamma, eps=eps,
156 |                         weight_decay=weight_decay, amsbound=amsbound)
157 |         super(AdaBoundW, self).__init__(params, defaults)
158 | 
159 |         self.base_lrs = list(map(lambda group: group['lr'], self.param_groups))
160 | 
161 |     def __setstate__(self, state):
162 |         super(AdaBoundW, self).__setstate__(state)
163 |         for group in self.param_groups:
164 |             group.setdefault('amsbound', False)
165 | 
166 |     def step(self, closure=None):
167 |         """Performs a single optimization step.
168 |         Arguments:
169 |             closure (callable, optional): A closure that reevaluates the model
170 |                 and returns the loss.
171 |         """
172 |         loss = None
173 |         if closure is not None:
174 |             loss = closure()
175 | 
176 |         for group, base_lr in zip(self.param_groups, self.base_lrs):
177 |             for p in group['params']:
178 |                 if p.grad is None:
179 |                     continue
180 |                 grad = p.grad.data
181 |                 if grad.is_sparse:
182 |                     raise RuntimeError(
183 |                         'Adam does not support sparse gradients, please consider SparseAdam instead')
184 |                 amsbound = group['amsbound']
185 | 
186 |                 state = self.state[p]
187 | 
188 |                 # State initialization
189 |                 if len(state) == 0:
190 |                     state['step'] = 0
191 |                     # Exponential moving average of gradient values
192 |                     state['exp_avg'] = torch.zeros_like(p.data)
193 |                     # Exponential moving average of squared gradient values
194 |                     state['exp_avg_sq'] = torch.zeros_like(p.data)
195 |                     if amsbound:
196 |                         # Maintains max of all exp. moving avg. of sq. grad. values
197 |                         state['max_exp_avg_sq'] = torch.zeros_like(p.data)
198 | 
199 |                 exp_avg, exp_avg_sq = state['exp_avg'], state['exp_avg_sq']
200 |                 if amsbound:
201 |                     max_exp_avg_sq = state['max_exp_avg_sq']
202 |                 beta1, beta2 = group['betas']
203 | 
204 |                 state['step'] += 1
205 | 
206 |                 # Decay the first and second moment running average coefficient
207 |                 exp_avg.mul_(beta1).add_(1 - beta1, grad)
208 |                 exp_avg_sq.mul_(beta2).addcmul_(1 - beta2, grad, grad)
209 |                 if amsbound:
210 |                     # Maintains the maximum of all 2nd moment running avg. till now
211 |                     torch.max(max_exp_avg_sq, exp_avg_sq, out=max_exp_avg_sq)
212 |                     # Use the max. for normalizing running avg. of gradient
213 |                     denom = max_exp_avg_sq.sqrt().add_(group['eps'])
214 |                 else:
215 |                     denom = exp_avg_sq.sqrt().add_(group['eps'])
216 | 
217 |                 bias_correction1 = 1 - beta1 ** state['step']
218 |                 bias_correction2 = 1 - beta2 ** state['step']
219 |                 step_size = group['lr'] * math.sqrt(bias_correction2) / bias_correction1
220 | 
221 |                 # Applies bounds on actual learning rate
222 |                 # lr_scheduler cannot affect final_lr, this is a workaround to apply lr decay
223 |                 final_lr = group['final_lr'] * group['lr'] / base_lr
224 |                 lower_bound = final_lr * (1 - 1 / (group['gamma'] * state['step'] + 1))
225 |                 upper_bound = final_lr * (1 + 1 / (group['gamma'] * state['step']))
226 |                 step_size = torch.full_like(denom, step_size)
227 |                 step_size.div_(denom).clamp_(lower_bound, upper_bound).mul_(exp_avg)
228 | 
229 |                 if group['weight_decay'] != 0:
230 |                     decayed_weights = torch.mul(p.data, group['weight_decay'])
231 |                     p.data.add_(-step_size)
232 |                     p.data.sub_(decayed_weights)
233 |                 else:
234 |                     p.data.add_(-step_size)
235 | 
236 |         return loss
237 | 


--------------------------------------------------------------------------------
/utils/gcp.sh:
--------------------------------------------------------------------------------
  1 | #!/usr/bin/env bash
  2 | 
  3 | # New VM
  4 | rm -rf sample_data yolov3 darknet apex coco cocoapi knife knifec
  5 | git clone https://github.com/ultralytics/yolov3
  6 | # git clone https://github.com/AlexeyAB/darknet && cd darknet && make GPU=1 CUDNN=1 CUDNN_HALF=1 OPENCV=0 && wget -c https://pjreddie.com/media/files/darknet53.conv.74 && cd ..
  7 | git clone https://github.com/NVIDIA/apex && cd apex && pip install -v --no-cache-dir --global-option="--cpp_ext" --global-option="--cuda_ext" . --user && cd .. && rm -rf apex
  8 | # git clone https://github.com/cocodataset/cocoapi && cd cocoapi/PythonAPI && make && cd ../.. && cp -r cocoapi/PythonAPI/pycocotools yolov3
  9 | sudo conda install -y -c conda-forge scikit-image tensorboard pycocotools
 10 | python3 -c "
 11 | from yolov3.utils.google_utils import gdrive_download
 12 | gdrive_download('1HaXkef9z6y5l4vUnCYgdmEAj61c6bfWO','coco.zip')"
 13 | sudo shutdown
 14 | 
 15 | # Re-clone
 16 | rm -rf yolov3  # Warning: remove existing
 17 | git clone https://github.com/ultralytics/yolov3 && cd yolov3 # master
 18 | # git clone -b test --depth 1 https://github.com/ultralytics/yolov3 test  # branch
 19 | python3 train.py --img-size 320 --weights weights/darknet53.conv.74 --epochs 27 --batch-size 64 --accumulate 1
 20 | 
 21 | # Train
 22 | python3 train.py
 23 | 
 24 | # Resume
 25 | python3 train.py --resume
 26 | 
 27 | # Detect
 28 | python3 detect.py
 29 | 
 30 | # Test
 31 | python3 test.py --save-json
 32 | 
 33 | # Evolve
 34 | for i in {0..500}
 35 | do
 36 |   python3 train.py --data data/coco.data --img-size 320 --epochs 1 --batch-size 64 --accumulate 1 --evolve --bucket yolov4
 37 | done
 38 | 
 39 | # Git pull
 40 | git pull https://github.com/ultralytics/yolov3  # master
 41 | git pull https://github.com/ultralytics/yolov3 test  # branch
 42 | 
 43 | # Test Darknet training
 44 | python3 test.py --weights ../darknet/backup/yolov3.backup
 45 | 
 46 | # Copy last.pt TO bucket
 47 | gsutil cp yolov3/weights/last1gpu.pt gs://ultralytics
 48 | 
 49 | # Copy last.pt FROM bucket
 50 | gsutil cp gs://ultralytics/last.pt yolov3/weights/last.pt
 51 | wget https://storage.googleapis.com/ultralytics/yolov3/last_v1_0.pt -O weights/last_v1_0.pt
 52 | wget https://storage.googleapis.com/ultralytics/yolov3/best_v1_0.pt -O weights/best_v1_0.pt
 53 | 
 54 | # Reproduce tutorials
 55 | rm results*.txt  # WARNING: removes existing results
 56 | python3 train.py --nosave --data data/coco_1img.data && mv results.txt results0r_1img.txt
 57 | python3 train.py --nosave --data data/coco_10img.data && mv results.txt results0r_10img.txt
 58 | python3 train.py --nosave --data data/coco_100img.data && mv results.txt results0r_100img.txt
 59 | # python3 train.py --nosave --data data/coco_100img.data --transfer && mv results.txt results3_100imgTL.txt
 60 | python3 -c "from utils import utils; utils.plot_results()"
 61 | # gsutil cp results*.txt gs://ultralytics
 62 | gsutil cp results.png gs://ultralytics
 63 | sudo shutdown
 64 | 
 65 | # Reproduce mAP
 66 | python3 test.py --save-json --img-size 608
 67 | python3 test.py --save-json --img-size 416
 68 | python3 test.py --save-json --img-size 320
 69 | sudo shutdown
 70 | 
 71 | # Benchmark script
 72 | git clone https://github.com/ultralytics/yolov3  # clone our repo
 73 | git clone https://github.com/NVIDIA/apex && cd apex && pip install -v --no-cache-dir --global-option="--cpp_ext" --global-option="--cuda_ext" . --user && cd .. && rm -rf apex  # install nvidia apex
 74 | python3 -c "from yolov3.utils.google_utils import gdrive_download; gdrive_download('1HaXkef9z6y5l4vUnCYgdmEAj61c6bfWO','coco.zip')"  # download coco dataset (20GB)
 75 | cd yolov3 && clear && python3 train.py --epochs 1  # run benchmark (~30 min)
 76 | 
 77 | # Unit tests
 78 | python3 detect.py  # detect 2 persons, 1 tie
 79 | python3 test.py --data data/coco_32img.data  # test mAP = 0.8
 80 | python3 train.py --data data/coco_32img.data --epochs 5 --nosave  # train 5 epochs
 81 | python3 train.py --data data/coco_1cls.data --epochs 5 --nosave  # train 5 epochs
 82 | python3 train.py --data data/coco_1img.data --epochs 5 --nosave  # train 5 epochs
 83 | 
 84 | # AlexyAB Darknet
 85 | gsutil cp -r gs://sm6/supermarket2 .  # dataset from bucket
 86 | rm -rf darknet && git clone https://github.com/AlexeyAB/darknet && cd darknet && wget -c https://pjreddie.com/media/files/darknet53.conv.74  # sudo apt install libopencv-dev && make
 87 | ./darknet detector calc_anchors data/coco_img64.data -num_of_clusters 9 -width 320 -height 320  # kmeans anchor calculation
 88 | ./darknet detector train ../supermarket2/supermarket2.data ../yolo_v3_spp_pan_scale.cfg darknet53.conv.74 -map -dont_show # train spp
 89 | ./darknet detector train ../yolov3/data/coco.data ../yolov3-spp.cfg darknet53.conv.74 -map -dont_show # train spp coco
 90 | 
 91 | ./darknet detector train data/coco.data ../yolov3-spp.cfg darknet53.conv.74 -map -dont_show # train spp
 92 | gsutil cp -r backup/*5000.weights gs://sm6/weights
 93 | sudo shutdown
 94 | 
 95 | 
 96 | ./darknet detector train ../supermarket2/supermarket2.data ../yolov3-tiny-sm2-1cls.cfg yolov3-tiny.conv.15 -map -dont_show # train tiny
 97 | ./darknet detector train ../supermarket2/supermarket2.data cfg/yolov3-spp-sm2-1cls.cfg backup/yolov3-spp-sm2-1cls_last.weights  # resume
 98 | python3 train.py --data ../supermarket2/supermarket2.data --cfg ../yolov3-spp-sm2-1cls.cfg --epochs 100 --num-workers 8 --img-size 320 --nosave  # train ultralytics
 99 | python3 test.py --data ../supermarket2/supermarket2.data --weights ../darknet/backup/yolov3-spp-sm2-1cls_5000.weights --cfg cfg/yolov3-spp-sm2-1cls.cfg  # test
100 | gsutil cp -r backup/*.weights gs://sm6/weights  # weights to bucket
101 | 
102 | python3 test.py --data ../supermarket2/supermarket2.data --weights weights/yolov3-spp-sm2-1cls_5000.weights --cfg ../yolov3-spp-sm2-1cls.cfg --img-size 320 --conf-thres 0.2  # test
103 | python3 test.py --data ../supermarket2/supermarket2.data --weights weights/yolov3-spp-sm2-1cls-scalexy_125_5000.weights --cfg ../yolov3-spp-sm2-1cls-scalexy_125.cfg --img-size 320 --conf-thres 0.2  # test
104 | python3 test.py --data ../supermarket2/supermarket2.data --weights weights/yolov3-spp-sm2-1cls-scalexy_150_5000.weights --cfg ../yolov3-spp-sm2-1cls-scalexy_150.cfg --img-size 320 --conf-thres 0.2  # test
105 | python3 test.py --data ../supermarket2/supermarket2.data --weights weights/yolov3-spp-sm2-1cls-scalexy_200_5000.weights --cfg ../yolov3-spp-sm2-1cls-scalexy_200.cfg --img-size 320 --conf-thres 0.2  # test
106 | python3 test.py --data ../supermarket2/supermarket2.data --weights ../darknet/backup/yolov3-spp-sm2-1cls-scalexy_variable_5000.weights --cfg ../yolov3-spp-sm2-1cls-scalexy_variable.cfg --img-size 320 --conf-thres 0.2  # test
107 | 
108 | python3 train.py --img-size 320 --epochs 27 --batch-size 64 --accumulate 1 --nosave --notest && python3 test.py --weights weights/last.pt --img-size 320 --save-json && sudo shutdown
109 | 
110 | # Debug/Development
111 | python3 train.py --data data/coco.data --img-size 320 --single-scale --batch-size 64 --accumulate 1 --epochs 1 --evolve --giou
112 | python3 test.py --weights weights/last.pt --cfg cfg/yolov3-spp.cfg --img-size 320
113 | 
114 | gsutil cp evolve.txt gs://ultralytics
115 | sudo shutdown
116 | 
117 | #Docker
118 | sudo docker kill $(sudo docker ps -q)
119 | sudo docker pull ultralytics/yolov3:v1
120 | sudo nvidia-docker run -it --ipc=host --mount type=bind,source="$(pwd)"/coco,target=/usr/src/coco ultralytics/yolov3:v1
121 | 
122 | clear
123 | while true
124 | do
125 |   python3 train.py --data data/coco.data --img-size 320 --batch-size 64 --accumulate 1 --evolve --epochs 1 --adam --bucket yolov4/adamdefaultpw_coco_1e --device 1
126 | done
127 | 
128 | python3 train.py --data data/coco.data --img-size 320 --batch-size 64 --accumulate 1 --epochs 1 --adam --device 1 --prebias
129 | while true; do python3 train.py --data data/coco.data --img-size 320 --batch-size 64 --accumulate 1 --evolve --epochs 1 --adam --bucket yolov4/adamdefaultpw_coco_1e; done
130 | 


--------------------------------------------------------------------------------
/utils/google_utils.py:
--------------------------------------------------------------------------------
 1 | # This file contains google utils: https://cloud.google.com/storage/docs/reference/libraries
 2 | # pip install --upgrade google-cloud-storage
 3 | 
 4 | import os
 5 | import time
 6 | 
 7 | 
 8 | # from google.cloud import storage
 9 | 
10 | 
11 | def gdrive_download(id='1HaXkef9z6y5l4vUnCYgdmEAj61c6bfWO', name='coco.zip'):
12 |     # https://gist.github.com/tanaikech/f0f2d122e05bf5f971611258c22c110f
13 |     # Downloads a file from Google Drive, accepting presented query
14 |     # from utils.google_utils import *; gdrive_download()
15 |     t = time.time()
16 | 
17 |     print('Downloading https://drive.google.com/uc?export=download&id=%s as %s... ' % (id, name), end='')
18 |     if os.path.exists(name):  # remove existing
19 |         os.remove(name)
20 | 
21 |     # Attempt large file download
22 |     s = ["curl -c ./cookie -s -L \"https://drive.google.com/uc?export=download&id=%s\" > /dev/null" % id,
23 |          "curl -Lb ./cookie -s \"https://drive.google.com/uc?export=download&confirm=`awk '/download/ {print $NF}' ./cookie`&id=%s\" -o %s" % (
24 |              id, name),
25 |          'rm ./cookie']
26 |     [os.system(x) for x in s]  # run commands
27 | 
28 |     # Attempt small file download
29 |     if not os.path.exists(name):  # file size < 40MB
30 |         s = 'curl -f -L -o %s https://drive.google.com/uc?export=download&id=%s' % (name, id)
31 |         os.system(s)
32 | 
33 |     # Unzip if archive
34 |     if name.endswith('.zip'):
35 |         print('unzipping... ', end='')
36 |         os.system('unzip -q %s' % name)  # unzip
37 |         os.remove(name)  # remove zip to free space
38 | 
39 |     print('Done (%.1fs)' % (time.time() - t))
40 | 
41 | 
42 | def upload_blob(bucket_name, source_file_name, destination_blob_name):
43 |     # Uploads a file to a bucket
44 |     # https://cloud.google.com/storage/docs/uploading-objects#storage-upload-object-python
45 | 
46 |     storage_client = storage.Client()
47 |     bucket = storage_client.get_bucket(bucket_name)
48 |     blob = bucket.blob(destination_blob_name)
49 | 
50 |     blob.upload_from_filename(source_file_name)
51 | 
52 |     print('File {} uploaded to {}.'.format(
53 |         source_file_name,
54 |         destination_blob_name))
55 | 
56 | 
57 | def download_blob(bucket_name, source_blob_name, destination_file_name):
58 |     # Uploads a blob from a bucket
59 |     storage_client = storage.Client()
60 |     bucket = storage_client.get_bucket(bucket_name)
61 |     blob = bucket.blob(source_blob_name)
62 | 
63 |     blob.download_to_filename(destination_file_name)
64 | 
65 |     print('Blob {} downloaded to {}.'.format(
66 |         source_blob_name,
67 |         destination_file_name))
68 | 


--------------------------------------------------------------------------------
/utils/parse_config.py:
--------------------------------------------------------------------------------
 1 | import numpy as np
 2 | import torch
 3 | 
 4 | def parse_model_cfg(path):
 5 |     # Parses the yolo-v3 layer configuration file and returns module definitions
 6 |     file = open(path, 'r')
 7 |     lines = file.read().split('\n')
 8 |     lines = [x for x in lines if x and not x.startswith('#')]
 9 |     lines = [x.rstrip().lstrip() for x in lines]  # get rid of fringe whitespaces
10 |     mdefs = []  # module definitions
11 |     for line in lines:
12 |         if line.startswith('['):  # This marks the start of a new block
13 |             mdefs.append({})
14 |             mdefs[-1]['type'] = line[1:-1].rstrip()
15 |             if mdefs[-1]['type'] == 'convolutional':
16 |                 mdefs[-1]['batch_normalize'] = 0  # pre-populate with zeros (may be overwritten later)
17 |         else:
18 |             key, val = line.split("=")
19 |             key = key.rstrip()
20 | 
21 |             if 'anchors' in key:
22 |                 mdefs[-1][key] = np.array([float(x) for x in val.split(',')]).reshape((-1, 2))  # np anchors
23 |             else:
24 |                 mdefs[-1][key] = val.strip()
25 | 
26 |     return mdefs
27 | 
28 | 
29 | def parse_data_cfg(path):
30 |     # Parses the data configuration file
31 |     options = dict()
32 |     with open(path, 'r') as fp:
33 |         lines = fp.readlines()
34 | 
35 |     for line in lines:
36 |         line = line.strip()
37 |         if line == '' or line.startswith('#'):
38 |             continue
39 |         key, val = line.split('=')
40 |         options[key.strip()] = val.strip()
41 | 
42 |     return options
43 | 
44 | 


--------------------------------------------------------------------------------
/utils/torch_utils.py:
--------------------------------------------------------------------------------
  1 | import os
  2 | import time
  3 | 
  4 | import torch
  5 | import torch.nn as nn
  6 | 
  7 | 
  8 | def init_seeds(seed=0):
  9 |     torch.manual_seed(seed)
 10 |     torch.cuda.manual_seed(seed)
 11 |     torch.cuda.manual_seed_all(seed)
 12 | 
 13 |     # Remove randomness (may be slower on Tesla GPUs) # https://pytorch.org/docs/stable/notes/randomness.html
 14 |     if seed == 0:
 15 |         torch.backends.cudnn.deterministic = True
 16 |         torch.backends.cudnn.benchmark = False
 17 | 
 18 | 
 19 | def select_device(device='', apex=False):
 20 |     # device = 'cpu' or '0' or '0,1,2,3'
 21 |     cpu_request = device.lower() == 'cpu'
 22 |     if device and not cpu_request:  # if device requested other than 'cpu'
 23 |         os.environ['CUDA_VISIBLE_DEVICES'] = device  # set environment variable
 24 |         assert torch.cuda.is_available(), 'CUDA unavailable, invalid device %s requested' % device  # check availablity
 25 | 
 26 |     cuda = False if cpu_request else torch.cuda.is_available()
 27 |     if cuda:
 28 |         c = 1024 ** 2  # bytes to MB
 29 |         ng = torch.cuda.device_count()
 30 |         x = [torch.cuda.get_device_properties(i) for i in range(ng)]
 31 |         cuda_str = 'Using CUDA ' + ('Apex ' if apex else '')  # apex for mixed precision https://github.com/NVIDIA/apex
 32 |         for i in range(0, ng):
 33 |             if i == 1:
 34 |                 cuda_str = ' ' * len(cuda_str)
 35 |             print("%sdevice%g _CudaDeviceProperties(name='%s', total_memory=%dMB)" %
 36 |                   (cuda_str, i, x[i].name, x[i].total_memory / c))
 37 |     else:
 38 |         print('Using CPU')
 39 | 
 40 |     print('')  # skip a line
 41 |     return torch.device('cuda:0' if cuda else 'cpu')
 42 | 
 43 | def time_synchronized():
 44 |     torch.cuda.synchronize() if torch.cuda.is_available() else None
 45 |     return time.time()
 46 | 
 47 | 
 48 | def is_parallel(model):
 49 |     # is model is parallel with DP or DDP
 50 |     return type(model) in (nn.parallel.DataParallel, nn.parallel.DistributedDataParallel)
 51 | 
 52 | def initialize_weights(model):
 53 |     for m in model.modules():
 54 |         t = type(m)
 55 |         if t is nn.Conv2d:
 56 |             pass  # nn.init.kaiming_normal_(m.weight, mode='fan_out', nonlinearity='relu')
 57 |         elif t is nn.BatchNorm2d:
 58 |             m.eps = 1e-3
 59 |             m.momentum = 0.03
 60 |         elif t in [nn.LeakyReLU, nn.ReLU, nn.ReLU6]:
 61 |             m.inplace = True
 62 | 
 63 | def fuse_conv_and_bn(conv, bn):
 64 |     # https://tehnokv.com/posts/fusing-batchnorm-and-conv/
 65 |     with torch.no_grad():
 66 |         # init
 67 |         fusedconv = torch.nn.Conv2d(conv.in_channels,
 68 |                                     conv.out_channels,
 69 |                                     kernel_size=conv.kernel_size,
 70 |                                     stride=conv.stride,
 71 |                                     padding=conv.padding,
 72 |                                     bias=True)
 73 | 
 74 |         # prepare filters
 75 |         w_conv = conv.weight.clone().view(conv.out_channels, -1)
 76 |         w_bn = torch.diag(bn.weight.div(torch.sqrt(bn.eps + bn.running_var)))
 77 |         fusedconv.weight.copy_(torch.mm(w_bn, w_conv).view(fusedconv.weight.size()))
 78 | 
 79 |         # prepare spatial bias
 80 |         if conv.bias is not None:
 81 |             b_conv = conv.bias
 82 |         else:
 83 |             b_conv = torch.zeros(conv.weight.size(0))
 84 |         b_bn = bn.bias - bn.weight.mul(bn.running_mean).div(torch.sqrt(bn.running_var + bn.eps))
 85 |         fusedconv.bias.copy_(b_conv + b_bn)
 86 | 
 87 |         return fusedconv
 88 | 
 89 | 
 90 | def model_info(model, report='summary'):
 91 |     # Plots a line-by-line description of a PyTorch model
 92 |     n_p = sum(x.numel() for x in model.parameters())  # number parameters
 93 |     n_g = sum(x.numel() for x in model.parameters() if x.requires_grad)  # number gradients
 94 |     if report is 'full':
 95 |         print('%5s %40s %9s %12s %20s %10s %10s' % ('layer', 'name', 'gradient', 'parameters', 'shape', 'mu', 'sigma'))
 96 |         for i, (name, p) in enumerate(model.named_parameters()):
 97 |             name = name.replace('module_list.', '')
 98 |             print('%5g %40s %9s %12g %20s %10.3g %10.3g' %
 99 |                   (i, name, p.requires_grad, p.numel(), list(p.shape), p.mean(), p.std()))
100 |     print('Model Summary: %g layers, %g parameters, %g gradients' % (len(list(model.parameters())), n_p, n_g))
101 | 


--------------------------------------------------------------------------------