├── .gitignore ├── LICENSE ├── README.md ├── _init_paths.py ├── cfgs ├── res101.yml ├── res101_ls.yml ├── res50.yml └── vgg16.yml ├── demo_global.py ├── imgs ├── architecture.png └── result.png ├── lib ├── datasets │ ├── VOCdevkit-matlab-wrapper │ │ ├── get_voc_opts.m │ │ ├── voc_eval.m │ │ └── xVOCap.m │ ├── __init__.py │ ├── bdd100k.py │ ├── cityscape.py │ ├── cityscape_car999.py │ ├── clipart.py │ ├── coco.py │ ├── config_dataset.py │ ├── ds_utils.py │ ├── factory.py │ ├── foggy_cityscape.py │ ├── imdb.py │ ├── pascal_voc.py │ ├── pascal_voc_cycleclipart.py │ ├── pascal_voc_cyclewater.py │ ├── pascal_voc_water.py │ ├── sim10k.py │ ├── sim10k_cycle.py │ ├── tools │ │ ├── list_all_images.py │ │ ├── mcg_munge.py │ │ └── multilabel_list.py │ ├── voc_eval.py │ ├── voc_eval_backup.py │ ├── voc_eval_modified.py │ └── water.py ├── make.sh ├── model │ ├── __init__.py │ ├── faster_rcnn │ │ ├── __init__.py │ │ ├── faster_rcnn_MEAA.py │ │ ├── networks.py │ │ ├── resnet_MEAA.py │ │ ├── resnet_dafrcnn.py │ │ └── vgg16_MEAA.py │ ├── nms │ │ ├── .gitignore │ │ ├── __init__.py │ │ ├── _ext │ │ │ ├── __init__.py │ │ │ └── nms │ │ │ │ └── __init__.py │ │ ├── build.py │ │ ├── make.sh │ │ ├── nms_cpu.py │ │ ├── nms_gpu.py │ │ ├── nms_kernel.cu │ │ ├── nms_wrapper.py │ │ └── src │ │ │ ├── nms_cuda.c │ │ │ ├── nms_cuda.h │ │ │ ├── nms_cuda_kernel.cu │ │ │ └── nms_cuda_kernel.h │ ├── roi_align │ │ ├── __init__.py │ │ ├── _ext │ │ │ ├── __init__.py │ │ │ └── roi_align │ │ │ │ └── __init__.py │ │ ├── build.py │ │ ├── functions │ │ │ ├── __init__.py │ │ │ └── roi_align.py │ │ ├── make.sh │ │ ├── modules │ │ │ ├── __init__.py │ │ │ └── roi_align.py │ │ └── src │ │ │ ├── roi_align.c │ │ │ ├── roi_align.h │ │ │ ├── roi_align_cuda.c │ │ │ ├── roi_align_cuda.h │ │ │ ├── roi_align_kernel.cu │ │ │ └── roi_align_kernel.h │ ├── roi_crop │ │ ├── __init__.py │ │ ├── _ext │ │ │ ├── __init__.py │ │ │ ├── crop_resize │ │ │ │ └── __init__.py │ │ │ └── roi_crop │ │ │ │ └── __init__.py │ │ ├── build.py │ │ ├── functions │ │ │ ├── __init__.py │ │ │ ├── crop_resize.py │ │ │ ├── gridgen.py │ │ │ └── roi_crop.py │ │ ├── make.sh │ │ ├── modules │ │ │ ├── __init__.py │ │ │ ├── gridgen.py │ │ │ └── roi_crop.py │ │ └── src │ │ │ ├── roi_crop.c │ │ │ ├── roi_crop.h │ │ │ ├── roi_crop_cuda.c │ │ │ ├── roi_crop_cuda.h │ │ │ ├── roi_crop_cuda_kernel.cu │ │ │ └── roi_crop_cuda_kernel.h │ ├── roi_pooling │ │ ├── __init__.py │ │ ├── _ext │ │ │ ├── __init__.py │ │ │ └── roi_pooling │ │ │ │ └── __init__.py │ │ ├── build.py │ │ ├── functions │ │ │ ├── __init__.py │ │ │ └── roi_pool.py │ │ ├── modules │ │ │ ├── __init__.py │ │ │ └── roi_pool.py │ │ └── src │ │ │ ├── roi_pooling.c │ │ │ ├── roi_pooling.h │ │ │ ├── roi_pooling_cuda.c │ │ │ ├── roi_pooling_cuda.h │ │ │ ├── roi_pooling_kernel.cu │ │ │ └── roi_pooling_kernel.h │ ├── rpn │ │ ├── __init__.py │ │ ├── anchor_target_layer.py │ │ ├── bbox_transform.py │ │ ├── generate_anchors.py │ │ ├── proposal_layer.py │ │ ├── proposal_target_layer_cascade.py │ │ └── rpn.py │ └── utils │ │ ├── .gitignore │ │ ├── __init__.py │ │ ├── bbox.pyx │ │ ├── blob.py │ │ ├── config.py │ │ ├── net_utils.py │ │ ├── parser_func.py │ │ └── parser_func_multi.py ├── pycocotools │ ├── UPSTREAM_REV │ ├── __init__.py │ ├── _mask.c │ ├── _mask.pyx │ ├── coco.py │ ├── cocoeval.py │ ├── license.txt │ ├── mask.py │ ├── maskApi.c │ └── maskApi.h ├── roi_data_layer │ ├── __init__.py │ ├── minibatch.py │ ├── roibatchLoader.py │ └── roidb.py └── setup.py ├── requirements.txt ├── test_net_MEAA.py └── trainval_net_MEAA.py /.gitignore: -------------------------------------------------------------------------------- 1 | *.pkl 2 | *.pth 3 | *.so 4 | *.pyc 5 | *.o 6 | -------------------------------------------------------------------------------- /LICENSE: -------------------------------------------------------------------------------- 1 | MIT License 2 | 3 | Copyright (c) 2020 BASIC Lab 4 | 5 | Permission is hereby granted, free of charge, to any person obtaining a copy 6 | of this software and associated documentation files (the "Software"), to deal 7 | in the Software without restriction, including without limitation the rights 8 | to use, copy, modify, merge, publish, distribute, sublicense, and/or sell 9 | copies of the Software, and to permit persons to whom the Software is 10 | furnished to do so, subject to the following conditions: 11 | 12 | The above copyright notice and this permission notice shall be included in all 13 | copies or substantial portions of the Software. 14 | 15 | THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR 16 | IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 17 | FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE 18 | AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER 19 | LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, 20 | OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE 21 | SOFTWARE. 22 | -------------------------------------------------------------------------------- /README.md: -------------------------------------------------------------------------------- 1 | # Domain-Adaptive Object Detection via Uncertainty-Aware Distribution Alignment 2 | 3 | [![PWC](https://img.shields.io/endpoint.svg?url=https://paperswithcode.com/badge/domain-adaptive-object-detection-via-1/weakly-supervised-object-detection-on-1)](https://paperswithcode.com/sota/weakly-supervised-object-detection-on-1?p=domain-adaptive-object-detection-via-1) 4 | [![PWC](https://img.shields.io/endpoint.svg?url=https://paperswithcode.com/badge/domain-adaptive-object-detection-via-1/weakly-supervised-object-detection-on-2)](https://paperswithcode.com/sota/weakly-supervised-object-detection-on-2?p=domain-adaptive-object-detection-via-1) 5 | 6 | 7 | Multi-level Entropy Attention Alignment (MEAA) is an end-to-end approach for unsupervised domain adaptation of object detector. Specifically, MEAA consists of two main components: 8 | 9 | (1) Local Uncertainty Attentional Alignment (LUAA) module to accelerate the model better perceiving structure-invariant objects of interest by utilizing information theory to measure the uncertainty of each local region via the entropy of the pixel-wise domain classifier 10 | 11 | (2) Multi-level Uncertainty-Aware Context Alignment (MUCA) module to enrich domain-invariant information of relevant objects based on the entropy of multi-level domain classifiers 12 | 13 | ![Overall architecture design](https://github.com/basiclab/DA-OD-MEAA-PyTorch/blob/main/imgs/architecture.png) 14 | 15 | 16 | 17 | ## Setup Introduction 18 | Follow [faster-rcnn repository](https://github.com/jwyang/faster-rcnn.pytorch) 19 | to setup the environment. When installing pytorch-faster-rcnn, you may encounter some issues. 20 | Many issues have been reported there to setup the environment. We used Pytorch 0.4.1 for this project. 21 | The different version of pytorch will cause some errors, which have to be handled based on each envirionment. 22 | 23 | ### Tested Hardwards & Softwares 24 | - GTX 1080 25 | - Pytorch 0.4.1 26 | - CUDA 9.2 27 | ``` 28 | conda install pytorch=0.4.1 torchvision==0.2.1 cuda92 -c pytorch 29 | ``` 30 | - Before training: 31 | ``` 32 | mkdir data 33 | cd lib 34 | sh make.sh (add -gencode arch=compute_70,code=sm_70" # added for GTX10XX) 35 | ``` 36 | 37 | - Note to set number of classes = 20 in lib/datasets/water.py 38 | - Tensorboard 39 | `tensorboard --logdir='your/path/here'` 40 | 41 | 42 | ### Data Preparation 43 | 44 | * **PASCAL_VOC 07+12**: Please follow the instructions in [py-faster-rcnn](https://github.com/rbgirshick/py-faster-rcnn#beyond-the-demo-installation-for-training-and-testing-models) to prepare VOC datasets. 45 | * **Clipart, WaterColor**: Dataset preparation instruction link [Cross Domain Detection ](https://github.com/naoto0804/cross-domain-detection/tree/master/datasets). 46 | * **Sim10k**: Website [Sim10k](https://fcav.engin.umich.edu/sim-dataset/) 47 | * **CitysScape, FoggyCityscape**: Download website [Cityscape](https://www.cityscapes-dataset.com/), see dataset preparation code in [DA-Faster RCNN](https://github.com/yuhuayc/da-faster-rcnn/tree/master/prepare_data) 48 | 49 | All codes are written to fit for the format of PASCAL_VOC. 50 | For example, the dataset [Sim10k](https://fcav.engin.umich.edu/sim-dataset/) is stored as follows. 51 | 52 | ``` 53 | $ cd Sim10k/VOC2012/ 54 | $ ls 55 | Annotations ImageSets JPEGImages 56 | $ cat ImageSets/Main/val.txt 57 | 3384827.jpg 58 | 3384828.jpg 59 | 3384829.jpg 60 | . 61 | . 62 | . 63 | ``` 64 | If you want to test the code on your own dataset, arange the dataset 65 | in the format of PASCAL, make dataset class in lib/datasets/. and add 66 | it to lib/datasets/factory.py, lib/datasets/config_dataset.py. Then, add the dataset option to lib/model/utils/parser_func.py. 67 | 68 | ### Data Path 69 | Write your dataset directories' paths in lib/datasets/config_dataset.py. 70 | 71 | ### Pretrained Model 72 | 73 | We used two models pre-trained on ImageNet in our experiments, VGG and ResNet101. You can download these two models from: 74 | 75 | * VGG16: [Dropbox](https://www.dropbox.com/s/s3brpk0bdq60nyb/vgg16_caffe.pth?dl=0), [VT Server](https://filebox.ece.vt.edu/~jw2yang/faster-rcnn/pretrained-base-models/vgg16_caffe.pth) 76 | 77 | * ResNet101: [Dropbox](https://www.dropbox.com/s/iev3tkbz5wyyuz9/resnet101_caffe.pth?dl=0), [VT Server](https://filebox.ece.vt.edu/~jw2yang/faster-rcnn/pretrained-base-models/resnet101_caffe.pth) 78 | 79 | Download them and write the path in __C.VGG_PATH and __C.RESNET_PATH at lib/model/utils/config.py. 80 | 81 | 82 | ## Train 83 | - Cityscapes --> Foggy_cityscapes 84 | ``` 85 | python trainval_net_MEAA.py --cuda --net vgg16 --dataset cityscape --dataset_t foggy_cityscape 86 | ``` 87 | ### use tensorboard 88 | ``` 89 | python trainval_net_MEAA.py --cuda --net vgg16 --dataset cityscape --dataset_t foggy_cityscape --use_tfb 90 | ``` 91 | --use_tfb will enable tensorboard to record training results 92 | 93 | ## Test 94 | - Cityscapes --> Foggy_cityscapes 95 | ``` 96 | python test_net_MEAA.py --cuda --net vgg16 --dataset foggy_cityscape --load_name models/vgg16/cityscape/*.pth 97 | ``` 98 | Our trained model for foggy_cityscape: 99 | https://drive.google.com/file/d/17pDu7mrxtx4cbpV2HNCGm2fzqCM1BZqd/view?usp=sharing 100 | 101 | - Results: 102 | 103 | ![command line output results](https://github.com/basiclab/DA-OD-MEAA-PyTorch/blob/main/imgs/result.png) 104 | 105 | ## Reminder 106 | For training "pascasl_voc_0712 -> water" results, since we only use 6 classes for evaluation. 107 | We need to calculate results manually. 108 | Just use chosen 6 classes to calculate mAP. 109 | 110 | ## Demo 111 | This function is under construction now. 112 | ``` 113 | python demo_global.py --net vgg16 --load_name models/vgg16/cityscape/*.pth --cuda --dataset cityscape 114 | ``` 115 | ## References 116 | 117 | ``` 118 | @inproceedings{10.1145/3394171.3413553, 119 | author = {Nguyen, Dang-Khoa and Tseng, Wei-Lun and Shuai, Hong-Han}, 120 | title = {Domain-Adaptive Object Detection via Uncertainty-Aware Distribution Alignment}, 121 | year = {2020}, 122 | isbn = {9781450379885}, 123 | publisher = {Association for Computing Machinery}, 124 | address = {New York, NY, USA}, 125 | url = {https://doi.org/10.1145/3394171.3413553}, 126 | doi = {10.1145/3394171.3413553}, 127 | } 128 | ``` 129 | -------------------------------------------------------------------------------- /_init_paths.py: -------------------------------------------------------------------------------- 1 | import os.path as osp 2 | import sys 3 | 4 | def add_path(path): 5 | if path not in sys.path: 6 | sys.path.insert(0, path) 7 | 8 | this_dir = osp.dirname(__file__) 9 | 10 | # Add lib to PYTHONPATH 11 | lib_path = osp.join(this_dir, 'lib') 12 | add_path(lib_path) 13 | 14 | coco_path = osp.join(this_dir, 'data', 'coco', 'PythonAPI') 15 | add_path(coco_path) 16 | -------------------------------------------------------------------------------- /cfgs/res101.yml: -------------------------------------------------------------------------------- 1 | EXP_DIR: res101 2 | TRAIN: 3 | HAS_RPN: True 4 | BBOX_NORMALIZE_TARGETS_PRECOMPUTED: True 5 | RPN_POSITIVE_OVERLAP: 0.7 6 | RPN_BATCHSIZE: 256 7 | PROPOSAL_METHOD: gt 8 | BG_THRESH_LO: 0.0 9 | DISPLAY: 20 10 | BATCH_SIZE: 128 11 | RPN_POST_NMS_TOP_N_TARGET: 128 12 | WEIGHT_DECAY: 0.0001 13 | DOUBLE_BIAS: False 14 | LEARNING_RATE: 0.001 15 | TEST: 16 | HAS_RPN: True 17 | POOLING_SIZE: 7 18 | POOLING_MODE: align 19 | CROP_RESIZE_WITH_MAX_POOL: False 20 | -------------------------------------------------------------------------------- /cfgs/res101_ls.yml: -------------------------------------------------------------------------------- 1 | EXP_DIR: res101 2 | TRAIN: 3 | HAS_RPN: True 4 | BBOX_NORMALIZE_TARGETS_PRECOMPUTED: True 5 | RPN_POSITIVE_OVERLAP: 0.7 6 | RPN_BATCHSIZE: 256 7 | PROPOSAL_METHOD: gt 8 | BG_THRESH_LO: 0.0 9 | DISPLAY: 20 10 | BATCH_SIZE: 128 11 | WEIGHT_DECAY: 0.0001 12 | SCALES: [800] 13 | DOUBLE_BIAS: False 14 | LEARNING_RATE: 0.001 15 | TEST: 16 | HAS_RPN: True 17 | SCALES: [800] 18 | MAX_SIZE: 1200 19 | RPN_POST_NMS_TOP_N: 1000 20 | POOLING_SIZE: 7 21 | POOLING_MODE: align 22 | CROP_RESIZE_WITH_MAX_POOL: False 23 | -------------------------------------------------------------------------------- /cfgs/res50.yml: -------------------------------------------------------------------------------- 1 | EXP_DIR: res50 2 | TRAIN: 3 | HAS_RPN: True 4 | # IMS_PER_BATCH: 1 5 | BBOX_NORMALIZE_TARGETS_PRECOMPUTED: True 6 | RPN_POSITIVE_OVERLAP: 0.7 7 | RPN_BATCHSIZE: 256 8 | PROPOSAL_METHOD: gt 9 | BG_THRESH_LO: 0.0 10 | DISPLAY: 20 11 | BATCH_SIZE: 256 12 | WEIGHT_DECAY: 0.0001 13 | DOUBLE_BIAS: False 14 | SNAPSHOT_PREFIX: res50_faster_rcnn 15 | TEST: 16 | HAS_RPN: True 17 | POOLING_MODE: crop 18 | -------------------------------------------------------------------------------- /cfgs/vgg16.yml: -------------------------------------------------------------------------------- 1 | EXP_DIR: vgg16 2 | TRAIN: 3 | HAS_RPN: True 4 | BBOX_NORMALIZE_TARGETS_PRECOMPUTED: True 5 | RPN_POSITIVE_OVERLAP: 0.7 6 | RPN_BATCHSIZE: 256 7 | PROPOSAL_METHOD: gt 8 | BG_THRESH_LO: 0.0 9 | BATCH_SIZE: 256 10 | RPN_POST_NMS_TOP_N_TARGET: 256 11 | LEARNING_RATE: 0.001 12 | TEST: 13 | HAS_RPN: True 14 | POOLING_MODE: align 15 | CROP_RESIZE_WITH_MAX_POOL: False 16 | -------------------------------------------------------------------------------- /imgs/architecture.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/basiclab/DA-OD-MEAA-PyTorch/ab590ddacdd06886fdbd56f7c4f59cd36167d100/imgs/architecture.png -------------------------------------------------------------------------------- /imgs/result.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/basiclab/DA-OD-MEAA-PyTorch/ab590ddacdd06886fdbd56f7c4f59cd36167d100/imgs/result.png -------------------------------------------------------------------------------- /lib/datasets/VOCdevkit-matlab-wrapper/get_voc_opts.m: -------------------------------------------------------------------------------- 1 | function VOCopts = get_voc_opts(path) 2 | 3 | tmp = pwd; 4 | cd(path); 5 | try 6 | addpath('VOCcode'); 7 | VOCinit; 8 | catch 9 | rmpath('VOCcode'); 10 | cd(tmp); 11 | error(sprintf('VOCcode directory not found under %s', path)); 12 | end 13 | rmpath('VOCcode'); 14 | cd(tmp); 15 | -------------------------------------------------------------------------------- /lib/datasets/VOCdevkit-matlab-wrapper/voc_eval.m: -------------------------------------------------------------------------------- 1 | function res = voc_eval(path, comp_id, test_set, output_dir) 2 | 3 | VOCopts = get_voc_opts(path); 4 | VOCopts.testset = test_set; 5 | 6 | for i = 1:length(VOCopts.classes) 7 | cls = VOCopts.classes{i}; 8 | res(i) = voc_eval_cls(cls, VOCopts, comp_id, output_dir); 9 | end 10 | 11 | fprintf('\n~~~~~~~~~~~~~~~~~~~~\n'); 12 | fprintf('Results:\n'); 13 | aps = [res(:).ap]'; 14 | fprintf('%.1f\n', aps * 100); 15 | fprintf('%.1f\n', mean(aps) * 100); 16 | fprintf('~~~~~~~~~~~~~~~~~~~~\n'); 17 | 18 | function res = voc_eval_cls(cls, VOCopts, comp_id, output_dir) 19 | 20 | test_set = VOCopts.testset; 21 | year = VOCopts.dataset(4:end); 22 | 23 | addpath(fullfile(VOCopts.datadir, 'VOCcode')); 24 | 25 | res_fn = sprintf(VOCopts.detrespath, comp_id, cls); 26 | 27 | recall = []; 28 | prec = []; 29 | ap = 0; 30 | ap_auc = 0; 31 | 32 | do_eval = (str2num(year) <= 2007) | ~strcmp(test_set, 'test'); 33 | if do_eval 34 | % Bug in VOCevaldet requires that tic has been called first 35 | tic; 36 | [recall, prec, ap] = VOCevaldet(VOCopts, comp_id, cls, true); 37 | ap_auc = xVOCap(recall, prec); 38 | 39 | % force plot limits 40 | ylim([0 1]); 41 | xlim([0 1]); 42 | 43 | print(gcf, '-djpeg', '-r0', ... 44 | [output_dir '/' cls '_pr.jpg']); 45 | end 46 | fprintf('!!! %s : %.4f %.4f\n', cls, ap, ap_auc); 47 | 48 | res.recall = recall; 49 | res.prec = prec; 50 | res.ap = ap; 51 | res.ap_auc = ap_auc; 52 | 53 | save([output_dir '/' cls '_pr.mat'], ... 54 | 'res', 'recall', 'prec', 'ap', 'ap_auc'); 55 | 56 | rmpath(fullfile(VOCopts.datadir, 'VOCcode')); 57 | -------------------------------------------------------------------------------- /lib/datasets/VOCdevkit-matlab-wrapper/xVOCap.m: -------------------------------------------------------------------------------- 1 | function ap = xVOCap(rec,prec) 2 | % From the PASCAL VOC 2011 devkit 3 | 4 | mrec=[0 ; rec ; 1]; 5 | mpre=[0 ; prec ; 0]; 6 | for i=numel(mpre)-1:-1:1 7 | mpre(i)=max(mpre(i),mpre(i+1)); 8 | end 9 | i=find(mrec(2:end)~=mrec(1:end-1))+1; 10 | ap=sum((mrec(i)-mrec(i-1)).*mpre(i)); 11 | -------------------------------------------------------------------------------- /lib/datasets/__init__.py: -------------------------------------------------------------------------------- 1 | # -------------------------------------------------------- 2 | # Fast R-CNN 3 | # Copyright (c) 2015 Microsoft 4 | # Licensed under The MIT License [see LICENSE for details] 5 | # Written by Ross Girshick 6 | # -------------------------------------------------------- 7 | -------------------------------------------------------------------------------- /lib/datasets/config_dataset.py: -------------------------------------------------------------------------------- 1 | from __future__ import absolute_import 2 | from __future__ import division 3 | from __future__ import print_function 4 | 5 | import os 6 | import os.path as osp 7 | import numpy as np 8 | # `pip install easydict` if you don't have it 9 | from easydict import EasyDict as edict 10 | 11 | __D = edict() 12 | # Consumers can get config by: 13 | # from fast_rcnn_config import cfg 14 | cfg_d = __D 15 | # 16 | # Training options 17 | #with regard to pascal, the directories under the path will be ./VOC2007, ./VOC2012" 18 | __D.PASCAL = "Datasets/VOCdevkit/VOCdevkit" 19 | __D.PASCALCLIP = "" 20 | __D.PASCALWATER = "/VOCdevkit" 21 | 22 | #For these datasets, the directories under the path will be Annotations ImageSets JPEGImages." 23 | __D.CLIPART = "Datasets/clipart" 24 | __D.WATER = "Datasets/watercolor" 25 | __D.SIM10K = "Datasets/sim10k/VOC2012" 26 | __D.CITYSCAPE_CAR = "/VOC2007" 27 | __D.CITYSCAPE = "Datasets/cityscapes/VOC2007" 28 | __D.FOGGYCITY = "Datasets/foggy_cityscapes/VOC2007" 29 | __D.BDD100K = "Datasets/bdd100k/VOC2007" 30 | 31 | def _merge_a_into_b(a, b): 32 | """Merge config dictionary a into config dictionary b, clobbering the 33 | options in b whenever they are also specified in a. 34 | """ 35 | if type(a) is not edict: 36 | return 37 | 38 | for k, v in a.items(): 39 | # a must specify keys that are in b 40 | if k not in b: 41 | raise KeyError('{} is not a valid config key'.format(k)) 42 | 43 | # the types must match, too 44 | old_type = type(b[k]) 45 | if old_type is not type(v): 46 | if isinstance(b[k], np.ndarray): 47 | v = np.array(v, dtype=b[k].dtype) 48 | else: 49 | raise ValueError(('Type mismatch ({} vs. {}) ' 50 | 'for config key: {}').format(type(b[k]), 51 | type(v), k)) 52 | 53 | # recursively merge dicts 54 | if type(v) is edict: 55 | try: 56 | _merge_a_into_b(a[k], b[k]) 57 | except: 58 | print(('Error under config key: {}'.format(k))) 59 | raise 60 | else: 61 | b[k] = v 62 | 63 | 64 | def cfg_from_file(filename): 65 | """Load a config file and merge it into the default options.""" 66 | import yaml 67 | with open(filename, 'r') as f: 68 | yaml_cfg = edict(yaml.load(f)) 69 | 70 | _merge_a_into_b(yaml_cfg, __D) 71 | 72 | 73 | def cfg_from_list(cfg_list): 74 | """Set config keys via list (e.g., from command line).""" 75 | from ast import literal_eval 76 | assert len(cfg_list) % 2 == 0 77 | for k, v in zip(cfg_list[0::2], cfg_list[1::2]): 78 | key_list = k.split('.') 79 | d = __D 80 | for subkey in key_list[:-1]: 81 | assert subkey in d 82 | d = d[subkey] 83 | subkey = key_list[-1] 84 | assert subkey in d 85 | try: 86 | value = literal_eval(v) 87 | except: 88 | # handle the case when v is a string literal 89 | value = v 90 | assert type(value) == type(d[subkey]), \ 91 | 'type {} does not match original type {}'.format( 92 | type(value), type(d[subkey])) 93 | d[subkey] = value 94 | -------------------------------------------------------------------------------- /lib/datasets/ds_utils.py: -------------------------------------------------------------------------------- 1 | # -------------------------------------------------------- 2 | # Fast/er R-CNN 3 | # Licensed under The MIT License [see LICENSE for details] 4 | # Written by Ross Girshick 5 | # -------------------------------------------------------- 6 | from __future__ import absolute_import 7 | from __future__ import division 8 | from __future__ import print_function 9 | 10 | import numpy as np 11 | 12 | 13 | def unique_boxes(boxes, scale=1.0): 14 | """Return indices of unique boxes.""" 15 | v = np.array([1, 1e3, 1e6, 1e9]) 16 | hashes = np.round(boxes * scale).dot(v) 17 | _, index = np.unique(hashes, return_index=True) 18 | return np.sort(index) 19 | 20 | 21 | def xywh_to_xyxy(boxes): 22 | """Convert [x y w h] box format to [x1 y1 x2 y2] format.""" 23 | return np.hstack((boxes[:, 0:2], boxes[:, 0:2] + boxes[:, 2:4] - 1)) 24 | 25 | 26 | def xyxy_to_xywh(boxes): 27 | """Convert [x1 y1 x2 y2] box format to [x y w h] format.""" 28 | return np.hstack((boxes[:, 0:2], boxes[:, 2:4] - boxes[:, 0:2] + 1)) 29 | 30 | 31 | def validate_boxes(boxes, width=0, height=0): 32 | """Check that a set of boxes are valid.""" 33 | x1 = boxes[:, 0] 34 | y1 = boxes[:, 1] 35 | x2 = boxes[:, 2] 36 | y2 = boxes[:, 3] 37 | assert (x1 >= 0).all() 38 | assert (y1 >= 0).all() 39 | assert (x2 >= x1).all() 40 | assert (y2 >= y1).all() 41 | assert (x2 < width).all() 42 | assert (y2 < height).all() 43 | 44 | 45 | def filter_small_boxes(boxes, min_size): 46 | w = boxes[:, 2] - boxes[:, 0] 47 | h = boxes[:, 3] - boxes[:, 1] 48 | keep = np.where((w >= min_size) & (h > min_size))[0] 49 | return keep 50 | -------------------------------------------------------------------------------- /lib/datasets/factory.py: -------------------------------------------------------------------------------- 1 | # -------------------------------------------------------- 2 | # Fast R-CNN 3 | # Copyright (c) 2015 Microsoft 4 | # Licensed under The MIT License [see LICENSE for details] 5 | # Written by Ross Girshick 6 | # -------------------------------------------------------- 7 | 8 | """Factory method for easily getting imdbs by name.""" 9 | from __future__ import absolute_import 10 | from __future__ import division 11 | from __future__ import print_function 12 | 13 | __sets = {} 14 | from datasets.pascal_voc import pascal_voc 15 | from datasets.pascal_voc_water import pascal_voc_water 16 | from datasets.pascal_voc_cyclewater import pascal_voc_cyclewater 17 | from datasets.pascal_voc_cycleclipart import pascal_voc_cycleclipart 18 | from datasets.sim10k import sim10k 19 | from datasets.water import water 20 | from datasets.clipart import clipart 21 | from datasets.sim10k_cycle import sim10k_cycle 22 | from datasets.cityscape import cityscape 23 | # from datasets.cityscape_car import cityscape_car 24 | from datasets.foggy_cityscape import foggy_cityscape 25 | from datasets.bdd100k import bdd100k 26 | 27 | import numpy as np 28 | for split in ['train', 'trainval','val','test']: 29 | name = 'cityscape_{}'.format(split) 30 | __sets[name] = (lambda split=split : cityscape(split)) 31 | 32 | # for split in ['train', 'trainval','val','test']: 33 | # name = 'cityscape_car_{}'.format(split) 34 | # __sets[name] = (lambda split=split : cityscape_car(split)) 35 | 36 | for split in ['train', 'test']: 37 | name = 'bdd100k_{}'.format(split) 38 | __sets[name] = (lambda split=split : bdd100k(split)) 39 | 40 | for split in ['train', 'trainval','test']: 41 | name = 'foggy_cityscape_{}'.format(split) 42 | __sets[name] = (lambda split=split : foggy_cityscape(split)) 43 | 44 | for split in ['train','val']: 45 | name = 'sim10k_{}'.format(split) 46 | __sets[name] = (lambda split=split : sim10k(split)) 47 | for split in ['train', 'val']: 48 | name = 'sim10k_cycle_{}'.format(split) 49 | __sets[name] = (lambda split=split: sim10k_cycle(split)) 50 | for year in ['2007', '2012']: 51 | for split in ['train', 'val', 'trainval', 'test']: 52 | name = 'voc_{}_{}'.format(year, split) 53 | __sets[name] = (lambda split=split, year=year: pascal_voc(split, year)) 54 | for year in ['2007', '2012']: 55 | for split in ['train', 'val', 'trainval', 'test']: 56 | name = 'voc_water_{}_{}'.format(year, split) 57 | __sets[name] = (lambda split=split, year=year: pascal_voc_water(split, year)) 58 | for year in ['2007', '2012']: 59 | for split in ['train', 'val', 'trainval', 'test']: 60 | name = 'voc_cycleclipart_{}_{}'.format(year, split) 61 | __sets[name] = (lambda split=split, year=year: pascal_voc_cycleclipart(split, year)) 62 | for year in ['2007', '2012']: 63 | for split in ['train', 'val', 'trainval', 'test']: 64 | name = 'voc_cyclewater_{}_{}'.format(year, split) 65 | __sets[name] = (lambda split=split, year=year: pascal_voc_cyclewater(split, year)) 66 | for year in ['2007']: 67 | for split in ['trainval', 'test']: 68 | name = 'clipart_{}'.format(split) 69 | __sets[name] = (lambda split=split : clipart(split,year)) 70 | for year in ['2007']: 71 | for split in ['train', 'test']: 72 | name = 'water_{}'.format(split) 73 | __sets[name] = (lambda split=split : water(split,year)) 74 | def get_imdb(name): 75 | """Get an imdb (image database) by name.""" 76 | if name not in __sets: 77 | raise KeyError('Unknown dataset: {}'.format(name)) 78 | return __sets[name]() 79 | 80 | 81 | def list_imdbs(): 82 | """List all registered imdbs.""" 83 | return list(__sets.keys()) 84 | -------------------------------------------------------------------------------- /lib/datasets/imdb.py: -------------------------------------------------------------------------------- 1 | # -------------------------------------------------------- 2 | # Fast R-CNN 3 | # Copyright (c) 2015 Microsoft 4 | # Licensed under The MIT License [see LICENSE for details] 5 | # Written by Ross Girshick and Xinlei Chen 6 | # -------------------------------------------------------- 7 | from __future__ import absolute_import 8 | from __future__ import division 9 | from __future__ import print_function 10 | 11 | import os 12 | import os.path as osp 13 | import PIL 14 | from model.utils.cython_bbox import bbox_overlaps 15 | import numpy as np 16 | import scipy.sparse 17 | from model.utils.config import cfg 18 | import pdb 19 | 20 | ROOT_DIR = osp.join(osp.dirname(__file__), '..', '..') 21 | 22 | class imdb(object): 23 | """Image database.""" 24 | 25 | def __init__(self, name, classes=None): 26 | self._name = name 27 | self._num_classes = 0 28 | if not classes: 29 | self._classes = [] 30 | else: 31 | self._classes = classes 32 | self._image_index = [] 33 | self._obj_proposer = 'gt' 34 | self._roidb = None 35 | self._roidb_handler = self.default_roidb 36 | # Use this dict for storing dataset specific config options 37 | self.config = {} 38 | 39 | @property 40 | def name(self): 41 | return self._name 42 | 43 | @property 44 | def num_classes(self): 45 | return len(self._classes) 46 | 47 | @property 48 | def classes(self): 49 | return self._classes 50 | 51 | @property 52 | def image_index(self): 53 | return self._image_index 54 | 55 | @property 56 | def roidb_handler(self): 57 | return self._roidb_handler 58 | 59 | @roidb_handler.setter 60 | def roidb_handler(self, val): 61 | self._roidb_handler = val 62 | 63 | def set_proposal_method(self, method): 64 | method = eval('self.' + method + '_roidb') 65 | self.roidb_handler = method 66 | 67 | @property 68 | def roidb(self): 69 | # A roidb is a list of dictionaries, each with the following keys: 70 | # boxes 71 | # gt_overlaps 72 | # gt_classes 73 | # flipped 74 | if self._roidb is not None: 75 | return self._roidb 76 | self._roidb = self.roidb_handler() 77 | return self._roidb 78 | 79 | @property 80 | def cache_path(self): 81 | cache_path = osp.abspath(osp.join(cfg.DATA_DIR, 'cache')) 82 | if not os.path.exists(cache_path): 83 | os.makedirs(cache_path) 84 | return cache_path 85 | 86 | @property 87 | def num_images(self): 88 | return len(self.image_index) 89 | 90 | def image_path_at(self, i): 91 | raise NotImplementedError 92 | 93 | def image_id_at(self, i): 94 | raise NotImplementedError 95 | 96 | def default_roidb(self): 97 | raise NotImplementedError 98 | 99 | def evaluate_detections(self, all_boxes, output_dir=None): 100 | """ 101 | all_boxes is a list of length number-of-classes. 102 | Each list element is a list of length number-of-images. 103 | Each of those list elements is either an empty list [] 104 | or a numpy array of detection. 105 | 106 | all_boxes[class][image] = [] or np.array of shape #dets x 5 107 | """ 108 | raise NotImplementedError 109 | 110 | def _get_widths(self): 111 | return [PIL.Image.open(self.image_path_at(i)).size[0] 112 | for i in range(self.num_images)] 113 | 114 | def append_flipped_images(self): 115 | num_images = self.num_images 116 | widths = self._get_widths() 117 | 118 | for i in range(num_images): 119 | boxes = self.roidb[i]['boxes'].copy() 120 | oldx1 = boxes[:, 0].copy() 121 | oldx2 = boxes[:, 2].copy() 122 | boxes[:, 0] = widths[i] - oldx2 - 1 123 | boxes[:, 2] = widths[i] - oldx1 - 1 124 | 125 | try: 126 | assert (boxes[:, 2] >= boxes[:, 0]).all() 127 | except: 128 | print('error') 129 | print(boxes[:, 2] >= boxes[:, 0]) 130 | print(boxes) 131 | print(widths[i]) 132 | if 'seg_map' in self.roidb[i].keys(): 133 | seg_map = self.roidb[i]['seg_map'][::-1, :] 134 | entry = {'boxes': boxes, 135 | 'gt_overlaps': self.roidb[i]['gt_overlaps'], 136 | 'gt_classes': self.roidb[i]['gt_classes'], 137 | 'flipped': True, 138 | 'seg_map':seg_map} 139 | else: 140 | entry = {'boxes': boxes, 141 | 'gt_overlaps': self.roidb[i]['gt_overlaps'], 142 | 'gt_classes': self.roidb[i]['gt_classes'], 143 | 'flipped': True} 144 | self.roidb.append(entry) 145 | self._image_index = self._image_index * 2 146 | 147 | def evaluate_recall(self, candidate_boxes=None, thresholds=None, 148 | area='all', limit=None): 149 | """Evaluate detection proposal recall metrics. 150 | 151 | Returns: 152 | results: dictionary of results with keys 153 | 'ar': average recall 154 | 'recalls': vector recalls at each IoU overlap threshold 155 | 'thresholds': vector of IoU overlap thresholds 156 | 'gt_overlaps': vector of all ground-truth overlaps 157 | """ 158 | # Record max overlap value for each gt box 159 | # Return vector of overlap values 160 | areas = {'all': 0, 'small': 1, 'medium': 2, 'large': 3, 161 | '96-128': 4, '128-256': 5, '256-512': 6, '512-inf': 7} 162 | area_ranges = [[0 ** 2, 1e5 ** 2], # all 163 | [0 ** 2, 32 ** 2], # small 164 | [32 ** 2, 96 ** 2], # medium 165 | [96 ** 2, 1e5 ** 2], # large 166 | [96 ** 2, 128 ** 2], # 96-128 167 | [128 ** 2, 256 ** 2], # 128-256 168 | [256 ** 2, 512 ** 2], # 256-512 169 | [512 ** 2, 1e5 ** 2], # 512-inf 170 | ] 171 | assert area in areas, 'unknown area range: {}'.format(area) 172 | area_range = area_ranges[areas[area]] 173 | gt_overlaps = np.zeros(0) 174 | num_pos = 0 175 | for i in range(self.num_images): 176 | # Checking for max_overlaps == 1 avoids including crowd annotations 177 | # (...pretty hacking :/) 178 | max_gt_overlaps = self.roidb[i]['gt_overlaps'].toarray().max(axis=1) 179 | gt_inds = np.where((self.roidb[i]['gt_classes'] > 0) & 180 | (max_gt_overlaps == 1))[0] 181 | gt_boxes = self.roidb[i]['boxes'][gt_inds, :] 182 | gt_areas = self.roidb[i]['seg_areas'][gt_inds] 183 | valid_gt_inds = np.where((gt_areas >= area_range[0]) & 184 | (gt_areas <= area_range[1]))[0] 185 | gt_boxes = gt_boxes[valid_gt_inds, :] 186 | num_pos += len(valid_gt_inds) 187 | 188 | if candidate_boxes is None: 189 | # If candidate_boxes is not supplied, the default is to use the 190 | # non-ground-truth boxes from this roidb 191 | non_gt_inds = np.where(self.roidb[i]['gt_classes'] == 0)[0] 192 | boxes = self.roidb[i]['boxes'][non_gt_inds, :] 193 | else: 194 | boxes = candidate_boxes[i] 195 | if boxes.shape[0] == 0: 196 | continue 197 | if limit is not None and boxes.shape[0] > limit: 198 | boxes = boxes[:limit, :] 199 | 200 | overlaps = bbox_overlaps(boxes.astype(np.float), 201 | gt_boxes.astype(np.float)) 202 | 203 | _gt_overlaps = np.zeros((gt_boxes.shape[0])) 204 | for j in range(gt_boxes.shape[0]): 205 | # find which proposal box maximally covers each gt box 206 | argmax_overlaps = overlaps.argmax(axis=0) 207 | # and get the iou amount of coverage for each gt box 208 | max_overlaps = overlaps.max(axis=0) 209 | # find which gt box is 'best' covered (i.e. 'best' = most iou) 210 | gt_ind = max_overlaps.argmax() 211 | gt_ovr = max_overlaps.max() 212 | assert (gt_ovr >= 0) 213 | # find the proposal box that covers the best covered gt box 214 | box_ind = argmax_overlaps[gt_ind] 215 | # record the iou coverage of this gt box 216 | _gt_overlaps[j] = overlaps[box_ind, gt_ind] 217 | assert (_gt_overlaps[j] == gt_ovr) 218 | # mark the proposal box and the gt box as used 219 | overlaps[box_ind, :] = -1 220 | overlaps[:, gt_ind] = -1 221 | # append recorded iou coverage level 222 | gt_overlaps = np.hstack((gt_overlaps, _gt_overlaps)) 223 | 224 | gt_overlaps = np.sort(gt_overlaps) 225 | if thresholds is None: 226 | step = 0.05 227 | thresholds = np.arange(0.5, 0.95 + 1e-5, step) 228 | recalls = np.zeros_like(thresholds) 229 | # compute recall for each iou threshold 230 | for i, t in enumerate(thresholds): 231 | recalls[i] = (gt_overlaps >= t).sum() / float(num_pos) 232 | # ar = 2 * np.trapz(recalls, thresholds) 233 | ar = recalls.mean() 234 | return {'ar': ar, 'recalls': recalls, 'thresholds': thresholds, 235 | 'gt_overlaps': gt_overlaps} 236 | 237 | def create_roidb_from_box_list(self, box_list, gt_roidb): 238 | assert len(box_list) == self.num_images, \ 239 | 'Number of boxes must match number of ground-truth images' 240 | roidb = [] 241 | for i in range(self.num_images): 242 | boxes = box_list[i] 243 | num_boxes = boxes.shape[0] 244 | overlaps = np.zeros((num_boxes, self.num_classes), dtype=np.float32) 245 | 246 | if gt_roidb is not None and gt_roidb[i]['boxes'].size > 0: 247 | gt_boxes = gt_roidb[i]['boxes'] 248 | gt_classes = gt_roidb[i]['gt_classes'] 249 | gt_overlaps = bbox_overlaps(boxes.astype(np.float), 250 | gt_boxes.astype(np.float)) 251 | argmaxes = gt_overlaps.argmax(axis=1) 252 | maxes = gt_overlaps.max(axis=1) 253 | I = np.where(maxes > 0)[0] 254 | overlaps[I, gt_classes[argmaxes[I]]] = maxes[I] 255 | 256 | overlaps = scipy.sparse.csr_matrix(overlaps) 257 | roidb.append({ 258 | 'boxes': boxes, 259 | 'gt_classes': np.zeros((num_boxes,), dtype=np.int32), 260 | 'gt_overlaps': overlaps, 261 | 'flipped': False, 262 | 'seg_areas': np.zeros((num_boxes,), dtype=np.float32), 263 | }) 264 | return roidb 265 | 266 | @staticmethod 267 | def merge_roidbs(a, b): 268 | assert len(a) == len(b) 269 | for i in range(len(a)): 270 | a[i]['boxes'] = np.vstack((a[i]['boxes'], b[i]['boxes'])) 271 | a[i]['gt_classes'] = np.hstack((a[i]['gt_classes'], 272 | b[i]['gt_classes'])) 273 | a[i]['gt_overlaps'] = scipy.sparse.vstack([a[i]['gt_overlaps'], 274 | b[i]['gt_overlaps']]) 275 | a[i]['seg_areas'] = np.hstack((a[i]['seg_areas'], 276 | b[i]['seg_areas'])) 277 | return a 278 | 279 | def competition_mode(self, on): 280 | """Turn competition mode on or off.""" 281 | pass 282 | -------------------------------------------------------------------------------- /lib/datasets/tools/list_all_images.py: -------------------------------------------------------------------------------- 1 | import os 2 | p_path = '/scratch4/keisaito/visda/train' 3 | dir_list = os.listdir(p_path) 4 | write_name = open('/scratch4/keisaito/visda/all_images_train.txt','w') 5 | for direc in dir_list: 6 | if not '.txt' in direc: 7 | files = os.listdir(os.path.join(p_path,direc)) 8 | for file in files: 9 | class_name = direc 10 | #if class_name == 'motorcycle': 11 | # class_name = 'motorbike' 12 | #if class_name == 'plant': 13 | # class_name = 'pottedplant' 14 | file_name = os.path.join(p_path,direc,file) 15 | write_name.write('%s %s\n'%(file_name,class_name)) 16 | 17 | -------------------------------------------------------------------------------- /lib/datasets/tools/mcg_munge.py: -------------------------------------------------------------------------------- 1 | from __future__ import print_function 2 | import os 3 | import sys 4 | 5 | """Hacky tool to convert file system layout of MCG boxes downloaded from 6 | http://www.eecs.berkeley.edu/Research/Projects/CS/vision/grouping/mcg/ 7 | so that it's consistent with those computed by Jan Hosang (see: 8 | http://www.mpi-inf.mpg.de/departments/computer-vision-and-multimodal- 9 | computing/research/object-recognition-and-scene-understanding/how- 10 | good-are-detection-proposals-really/) 11 | 12 | NB: Boxes from the MCG website are in (y1, x1, y2, x2) order. 13 | Boxes from Hosang et al. are in (x1, y1, x2, y2) order. 14 | """ 15 | 16 | def munge(src_dir): 17 | # stored as: ./MCG-COCO-val2014-boxes/COCO_val2014_000000193401.mat 18 | # want: ./MCG/mat/COCO_val2014_0/COCO_val2014_000000141/COCO_val2014_000000141334.mat 19 | 20 | files = os.listdir(src_dir) 21 | for fn in files: 22 | base, ext = os.path.splitext(fn) 23 | # first 14 chars / first 22 chars / all chars + .mat 24 | # COCO_val2014_0/COCO_val2014_000000447/COCO_val2014_000000447991.mat 25 | first = base[:14] 26 | second = base[:22] 27 | dst_dir = os.path.join('MCG', 'mat', first, second) 28 | if not os.path.exists(dst_dir): 29 | os.makedirs(dst_dir) 30 | src = os.path.join(src_dir, fn) 31 | dst = os.path.join(dst_dir, fn) 32 | print('MV: {} -> {}'.format(src, dst)) 33 | os.rename(src, dst) 34 | 35 | if __name__ == '__main__': 36 | # src_dir should look something like: 37 | # src_dir = 'MCG-COCO-val2014-boxes' 38 | src_dir = sys.argv[1] 39 | munge(src_dir) 40 | -------------------------------------------------------------------------------- /lib/datasets/tools/multilabel_list.py: -------------------------------------------------------------------------------- 1 | import os 2 | import xml.etree.ElementTree as ET 3 | import sys 4 | argvs = sys.argv 5 | 6 | def load_image_set_index(ref): 7 | """ 8 | Load the indexes listed in this dataset's image set file. 9 | """ 10 | # Example path to image set file: 11 | # self._devkit_path + /VOCdevkit2007/VOC2007/ImageSets/Main/val.txt 12 | image_set_file = os.path.join(ref) 13 | assert os.path.exists(image_set_file), \ 14 | 'Path does not exist: {}'.format(image_set_file) 15 | with open(image_set_file) as f: 16 | image_index = [x.strip() for x in f.readlines()] 17 | return image_index 18 | 19 | def load_pascal_annotation(ref_path, index): 20 | """ 21 | Load image and bounding boxes info from XML file in the PASCAL VOC 22 | format. 23 | """ 24 | filename = os.path.join(ref_path, 'Annotations', index + '.xml') 25 | tree = ET.parse(filename) 26 | objs = tree.findall('object') 27 | obj_list = [] 28 | for ix, obj in enumerate(objs): 29 | cls = obj.find('name').text.lower().strip() 30 | obj_list.append(cls) 31 | return list(set(obj_list)) 32 | 33 | indexes = load_image_set_index(argvs[1]) 34 | images_list = open(argvs[3],'w') 35 | for index in indexes: 36 | objs = load_pascal_annotation(argvs[2],index) 37 | write_word = os.path.join('/research/masaito/detection_dataset/VOCdevkit/VOC2007/JPEGImages', index + '.jpg' + ' ') 38 | for name in objs: 39 | write_word = write_word + name + ' ' 40 | images_list.write(write_word + '\n') 41 | 42 | -------------------------------------------------------------------------------- /lib/make.sh: -------------------------------------------------------------------------------- 1 | #!/usr/bin/env bash 2 | export PATH=/usr/local/cuda-9.2/bin${PATH:+:${PATH}} 3 | export CPATH=/usr/local/cuda-9.2/include${CPATH:+:${CPATH}} 4 | export LD_LIBRARY_PATH=/usr/local/cuda-9.2/lib64${LD_LIBRARY_PATH:+:${LD_LIBRARY_PATH}} 5 | # CUDA_PATH=/usr/local/cuda/ 6 | 7 | export CUDA_PATH=/usr/local/cuda-9.2/ 8 | #You may also want to ad the following 9 | #export C_INCLUDE_PATH=/opt/cuda/include 10 | 11 | export CXXFLAGS="-std=c++11" 12 | export CFLAGS="-std=c99" 13 | 14 | python setup.py build_ext --inplace 15 | rm -rf build 16 | 17 | CUDA_ARCH="-gencode arch=compute_30,code=sm_30 \ 18 | -gencode arch=compute_35,code=sm_35 \ 19 | -gencode arch=compute_50,code=sm_50 \ 20 | -gencode arch=compute_52,code=sm_52 \ 21 | -gencode arch=compute_60,code=sm_60 \ 22 | -gencode arch=compute_61,code=sm_61 \ 23 | -gencode arch=compute_70,code=sm_70 " 24 | 25 | # compile NMS 26 | cd model/nms/src 27 | echo "Compiling nms kernels by nvcc..." 28 | nvcc -c -o nms_cuda_kernel.cu.o nms_cuda_kernel.cu \ 29 | -D GOOGLE_CUDA=1 -x cu -Xcompiler -fPIC $CUDA_ARCH 30 | 31 | cd ../ 32 | python build.py 33 | 34 | # compile roi_pooling 35 | cd ../../ 36 | cd model/roi_pooling/src 37 | echo "Compiling roi pooling kernels by nvcc..." 38 | nvcc -c -o roi_pooling.cu.o roi_pooling_kernel.cu \ 39 | -D GOOGLE_CUDA=1 -x cu -Xcompiler -fPIC $CUDA_ARCH 40 | cd ../ 41 | python build.py 42 | 43 | # compile roi_align 44 | cd ../../ 45 | cd model/roi_align/src 46 | echo "Compiling roi align kernels by nvcc..." 47 | nvcc -c -o roi_align_kernel.cu.o roi_align_kernel.cu \ 48 | -D GOOGLE_CUDA=1 -x cu -Xcompiler -fPIC $CUDA_ARCH 49 | cd ../ 50 | python build.py 51 | 52 | # compile roi_crop 53 | cd ../../ 54 | cd model/roi_crop/src 55 | echo "Compiling roi crop kernels by nvcc..." 56 | nvcc -c -o roi_crop_cuda_kernel.cu.o roi_crop_cuda_kernel.cu \ 57 | -D GOOGLE_CUDA=1 -x cu -Xcompiler -fPIC $CUDA_ARCH 58 | cd ../ 59 | python build.py 60 | -------------------------------------------------------------------------------- /lib/model/__init__.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/basiclab/DA-OD-MEAA-PyTorch/ab590ddacdd06886fdbd56f7c4f59cd36167d100/lib/model/__init__.py -------------------------------------------------------------------------------- /lib/model/faster_rcnn/__init__.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/basiclab/DA-OD-MEAA-PyTorch/ab590ddacdd06886fdbd56f7c4f59cd36167d100/lib/model/faster_rcnn/__init__.py -------------------------------------------------------------------------------- /lib/model/nms/.gitignore: -------------------------------------------------------------------------------- 1 | *.c 2 | *.cpp 3 | *.so 4 | -------------------------------------------------------------------------------- /lib/model/nms/__init__.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/basiclab/DA-OD-MEAA-PyTorch/ab590ddacdd06886fdbd56f7c4f59cd36167d100/lib/model/nms/__init__.py -------------------------------------------------------------------------------- /lib/model/nms/_ext/__init__.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/basiclab/DA-OD-MEAA-PyTorch/ab590ddacdd06886fdbd56f7c4f59cd36167d100/lib/model/nms/_ext/__init__.py -------------------------------------------------------------------------------- /lib/model/nms/_ext/nms/__init__.py: -------------------------------------------------------------------------------- 1 | 2 | from torch.utils.ffi import _wrap_function 3 | from ._nms import lib as _lib, ffi as _ffi 4 | 5 | __all__ = [] 6 | def _import_symbols(locals): 7 | for symbol in dir(_lib): 8 | fn = getattr(_lib, symbol) 9 | if callable(fn): 10 | locals[symbol] = _wrap_function(fn, _ffi) 11 | else: 12 | locals[symbol] = fn 13 | __all__.append(symbol) 14 | 15 | _import_symbols(locals()) 16 | -------------------------------------------------------------------------------- /lib/model/nms/build.py: -------------------------------------------------------------------------------- 1 | from __future__ import print_function 2 | import os 3 | import torch 4 | from torch.utils.ffi import create_extension 5 | 6 | #this_file = os.path.dirname(__file__) 7 | 8 | sources = [] 9 | headers = [] 10 | defines = [] 11 | with_cuda = False 12 | 13 | if torch.cuda.is_available(): 14 | print('Including CUDA code.') 15 | sources += ['src/nms_cuda.c'] 16 | headers += ['src/nms_cuda.h'] 17 | defines += [('WITH_CUDA', None)] 18 | with_cuda = True 19 | 20 | this_file = os.path.dirname(os.path.realpath(__file__)) 21 | print(this_file) 22 | extra_objects = ['src/nms_cuda_kernel.cu.o'] 23 | extra_objects = [os.path.join(this_file, fname) for fname in extra_objects] 24 | print(extra_objects) 25 | 26 | ffi = create_extension( 27 | '_ext.nms', 28 | headers=headers, 29 | sources=sources, 30 | define_macros=defines, 31 | relative_to=__file__, 32 | with_cuda=with_cuda, 33 | extra_objects=extra_objects 34 | ) 35 | 36 | if __name__ == '__main__': 37 | ffi.build() 38 | -------------------------------------------------------------------------------- /lib/model/nms/make.sh: -------------------------------------------------------------------------------- 1 | #!/usr/bin/env bash 2 | 3 | # CUDA_PATH=/usr/local/cuda/ 4 | 5 | cd src 6 | echo "Compiling stnm kernels by nvcc..." 7 | nvcc -c -o nms_cuda_kernel.cu.o nms_cuda_kernel.cu -x cu -Xcompiler -fPIC -arch=sm_52 8 | 9 | cd ../ 10 | python build.py 11 | -------------------------------------------------------------------------------- /lib/model/nms/nms_cpu.py: -------------------------------------------------------------------------------- 1 | from __future__ import absolute_import 2 | 3 | import numpy as np 4 | import torch 5 | 6 | def nms_cpu(dets, thresh): 7 | dets = dets.numpy() 8 | x1 = dets[:, 0] 9 | y1 = dets[:, 1] 10 | x2 = dets[:, 2] 11 | y2 = dets[:, 3] 12 | scores = dets[:, 4] 13 | 14 | areas = (x2 - x1 + 1) * (y2 - y1 + 1) 15 | order = scores.argsort()[::-1] 16 | 17 | keep = [] 18 | while order.size > 0: 19 | i = order.item(0) 20 | keep.append(i) 21 | xx1 = np.maximum(x1[i], x1[order[1:]]) 22 | yy1 = np.maximum(y1[i], y1[order[1:]]) 23 | xx2 = np.minimum(x2[i], x2[order[1:]]) 24 | yy2 = np.minimum(y2[i], y2[order[1:]]) 25 | 26 | w = np.maximum(0.0, xx2 - xx1 + 1) 27 | h = np.maximum(0.0, yy2 - yy1 + 1) 28 | inter = w * h 29 | ovr = inter / (areas[i] + areas[order[1:]] - inter) 30 | 31 | inds = np.where(ovr <= thresh)[0] 32 | order = order[inds + 1] 33 | 34 | return torch.IntTensor(keep) 35 | -------------------------------------------------------------------------------- /lib/model/nms/nms_gpu.py: -------------------------------------------------------------------------------- 1 | from __future__ import absolute_import 2 | import torch 3 | import numpy as np 4 | from ._ext import nms 5 | import pdb 6 | 7 | def nms_gpu(dets, thresh): 8 | keep = dets.new(dets.size(0), 1).zero_().int() 9 | num_out = dets.new(1).zero_().int() 10 | nms.nms_cuda(keep, dets, num_out, thresh) 11 | keep = keep[:num_out[0]] 12 | return keep 13 | -------------------------------------------------------------------------------- /lib/model/nms/nms_kernel.cu: -------------------------------------------------------------------------------- 1 | // ------------------------------------------------------------------ 2 | // Faster R-CNN 3 | // Copyright (c) 2015 Microsoft 4 | // Licensed under The MIT License [see fast-rcnn/LICENSE for details] 5 | // Written by Shaoqing Ren 6 | // ------------------------------------------------------------------ 7 | 8 | #include "gpu_nms.hpp" 9 | #include 10 | #include 11 | 12 | #define CUDA_CHECK(condition) \ 13 | /* Code block avoids redefinition of cudaError_t error */ \ 14 | do { \ 15 | cudaError_t error = condition; \ 16 | if (error != cudaSuccess) { \ 17 | std::cout << cudaGetErrorString(error) << std::endl; \ 18 | } \ 19 | } while (0) 20 | 21 | #define DIVUP(m,n) ((m) / (n) + ((m) % (n) > 0)) 22 | int const threadsPerBlock = sizeof(unsigned long long) * 8; 23 | 24 | __device__ inline float devIoU(float const * const a, float const * const b) { 25 | float left = max(a[0], b[0]), right = min(a[2], b[2]); 26 | float top = max(a[1], b[1]), bottom = min(a[3], b[3]); 27 | float width = max(right - left + 1, 0.f), height = max(bottom - top + 1, 0.f); 28 | float interS = width * height; 29 | float Sa = (a[2] - a[0] + 1) * (a[3] - a[1] + 1); 30 | float Sb = (b[2] - b[0] + 1) * (b[3] - b[1] + 1); 31 | return interS / (Sa + Sb - interS); 32 | } 33 | 34 | __global__ void nms_kernel(const int n_boxes, const float nms_overlap_thresh, 35 | const float *dev_boxes, unsigned long long *dev_mask) { 36 | const int row_start = blockIdx.y; 37 | const int col_start = blockIdx.x; 38 | 39 | // if (row_start > col_start) return; 40 | 41 | const int row_size = 42 | min(n_boxes - row_start * threadsPerBlock, threadsPerBlock); 43 | const int col_size = 44 | min(n_boxes - col_start * threadsPerBlock, threadsPerBlock); 45 | 46 | __shared__ float block_boxes[threadsPerBlock * 5]; 47 | if (threadIdx.x < col_size) { 48 | block_boxes[threadIdx.x * 5 + 0] = 49 | dev_boxes[(threadsPerBlock * col_start + threadIdx.x) * 5 + 0]; 50 | block_boxes[threadIdx.x * 5 + 1] = 51 | dev_boxes[(threadsPerBlock * col_start + threadIdx.x) * 5 + 1]; 52 | block_boxes[threadIdx.x * 5 + 2] = 53 | dev_boxes[(threadsPerBlock * col_start + threadIdx.x) * 5 + 2]; 54 | block_boxes[threadIdx.x * 5 + 3] = 55 | dev_boxes[(threadsPerBlock * col_start + threadIdx.x) * 5 + 3]; 56 | block_boxes[threadIdx.x * 5 + 4] = 57 | dev_boxes[(threadsPerBlock * col_start + threadIdx.x) * 5 + 4]; 58 | } 59 | __syncthreads(); 60 | 61 | if (threadIdx.x < row_size) { 62 | const int cur_box_idx = threadsPerBlock * row_start + threadIdx.x; 63 | const float *cur_box = dev_boxes + cur_box_idx * 5; 64 | int i = 0; 65 | unsigned long long t = 0; 66 | int start = 0; 67 | if (row_start == col_start) { 68 | start = threadIdx.x + 1; 69 | } 70 | for (i = start; i < col_size; i++) { 71 | if (devIoU(cur_box, block_boxes + i * 5) > nms_overlap_thresh) { 72 | t |= 1ULL << i; 73 | } 74 | } 75 | const int col_blocks = DIVUP(n_boxes, threadsPerBlock); 76 | dev_mask[cur_box_idx * col_blocks + col_start] = t; 77 | } 78 | } 79 | 80 | void _set_device(int device_id) { 81 | int current_device; 82 | CUDA_CHECK(cudaGetDevice(¤t_device)); 83 | if (current_device == device_id) { 84 | return; 85 | } 86 | // The call to cudaSetDevice must come before any calls to Get, which 87 | // may perform initialization using the GPU. 88 | CUDA_CHECK(cudaSetDevice(device_id)); 89 | } 90 | 91 | void _nms(int* keep_out, int* num_out, const float* boxes_host, int boxes_num, 92 | int boxes_dim, float nms_overlap_thresh, int device_id) { 93 | _set_device(device_id); 94 | 95 | float* boxes_dev = NULL; 96 | unsigned long long* mask_dev = NULL; 97 | 98 | const int col_blocks = DIVUP(boxes_num, threadsPerBlock); 99 | 100 | CUDA_CHECK(cudaMalloc(&boxes_dev, 101 | boxes_num * boxes_dim * sizeof(float))); 102 | CUDA_CHECK(cudaMemcpy(boxes_dev, 103 | boxes_host, 104 | boxes_num * boxes_dim * sizeof(float), 105 | cudaMemcpyHostToDevice)); 106 | 107 | CUDA_CHECK(cudaMalloc(&mask_dev, 108 | boxes_num * col_blocks * sizeof(unsigned long long))); 109 | 110 | dim3 blocks(DIVUP(boxes_num, threadsPerBlock), 111 | DIVUP(boxes_num, threadsPerBlock)); 112 | dim3 threads(threadsPerBlock); 113 | nms_kernel<<>>(boxes_num, 114 | nms_overlap_thresh, 115 | boxes_dev, 116 | mask_dev); 117 | 118 | std::vector mask_host(boxes_num * col_blocks); 119 | CUDA_CHECK(cudaMemcpy(&mask_host[0], 120 | mask_dev, 121 | sizeof(unsigned long long) * boxes_num * col_blocks, 122 | cudaMemcpyDeviceToHost)); 123 | 124 | std::vector remv(col_blocks); 125 | memset(&remv[0], 0, sizeof(unsigned long long) * col_blocks); 126 | 127 | int num_to_keep = 0; 128 | for (int i = 0; i < boxes_num; i++) { 129 | int nblock = i / threadsPerBlock; 130 | int inblock = i % threadsPerBlock; 131 | 132 | if (!(remv[nblock] & (1ULL << inblock))) { 133 | keep_out[num_to_keep++] = i; 134 | unsigned long long *p = &mask_host[0] + i * col_blocks; 135 | for (int j = nblock; j < col_blocks; j++) { 136 | remv[j] |= p[j]; 137 | } 138 | } 139 | } 140 | *num_out = num_to_keep; 141 | 142 | CUDA_CHECK(cudaFree(boxes_dev)); 143 | CUDA_CHECK(cudaFree(mask_dev)); 144 | } 145 | -------------------------------------------------------------------------------- /lib/model/nms/nms_wrapper.py: -------------------------------------------------------------------------------- 1 | # -------------------------------------------------------- 2 | # Fast R-CNN 3 | # Copyright (c) 2015 Microsoft 4 | # Licensed under The MIT License [see LICENSE for details] 5 | # Written by Ross Girshick 6 | # -------------------------------------------------------- 7 | import torch 8 | from model.utils.config import cfg 9 | if torch.cuda.is_available(): 10 | from model.nms.nms_gpu import nms_gpu 11 | from model.nms.nms_cpu import nms_cpu 12 | 13 | def nms(dets, thresh, force_cpu=False): 14 | """Dispatch to either CPU or GPU NMS implementations.""" 15 | if dets.shape[0] == 0: 16 | return [] 17 | # ---numpy version--- 18 | # original: return gpu_nms(dets, thresh, device_id=cfg.GPU_ID) 19 | # ---pytorch version--- 20 | 21 | return nms_gpu(dets, thresh) if force_cpu == False else nms_cpu(dets, thresh) 22 | -------------------------------------------------------------------------------- /lib/model/nms/src/nms_cuda.c: -------------------------------------------------------------------------------- 1 | #include 2 | #include 3 | #include "nms_cuda_kernel.h" 4 | 5 | // this symbol will be resolved automatically from PyTorch libs 6 | extern THCState *state; 7 | 8 | int nms_cuda(THCudaIntTensor *keep_out, THCudaTensor *boxes_host, 9 | THCudaIntTensor *num_out, float nms_overlap_thresh) { 10 | 11 | nms_cuda_compute(THCudaIntTensor_data(state, keep_out), 12 | THCudaIntTensor_data(state, num_out), 13 | THCudaTensor_data(state, boxes_host), 14 | THCudaTensor_size(state, boxes_host, 0), 15 | THCudaTensor_size(state, boxes_host, 1), 16 | nms_overlap_thresh); 17 | 18 | return 1; 19 | } 20 | -------------------------------------------------------------------------------- /lib/model/nms/src/nms_cuda.h: -------------------------------------------------------------------------------- 1 | // int nms_cuda(THCudaTensor *keep_out, THCudaTensor *num_out, 2 | // THCudaTensor *boxes_host, THCudaTensor *nms_overlap_thresh); 3 | 4 | int nms_cuda(THCudaIntTensor *keep_out, THCudaTensor *boxes_host, 5 | THCudaIntTensor *num_out, float nms_overlap_thresh); 6 | -------------------------------------------------------------------------------- /lib/model/nms/src/nms_cuda_kernel.cu: -------------------------------------------------------------------------------- 1 | // ------------------------------------------------------------------ 2 | // Faster R-CNN 3 | // Copyright (c) 2015 Microsoft 4 | // Licensed under The MIT License [see fast-rcnn/LICENSE for details] 5 | // Written by Shaoqing Ren 6 | // ------------------------------------------------------------------ 7 | 8 | #include 9 | #include 10 | #include 11 | #include 12 | #include "nms_cuda_kernel.h" 13 | 14 | #define CUDA_WARN(XXX) \ 15 | do { if (XXX != cudaSuccess) std::cout << "CUDA Error: " << \ 16 | cudaGetErrorString(XXX) << ", at line " << __LINE__ \ 17 | << std::endl; cudaDeviceSynchronize(); } while (0) 18 | 19 | #define CUDA_CHECK(condition) \ 20 | /* Code block avoids redefinition of cudaError_t error */ \ 21 | do { \ 22 | cudaError_t error = condition; \ 23 | if (error != cudaSuccess) { \ 24 | std::cout << cudaGetErrorString(error) << std::endl; \ 25 | } \ 26 | } while (0) 27 | 28 | #define DIVUP(m,n) ((m) / (n) + ((m) % (n) > 0)) 29 | int const threadsPerBlock = sizeof(unsigned long long) * 8; 30 | 31 | __device__ inline float devIoU(float const * const a, float const * const b) { 32 | float left = max(a[0], b[0]), right = min(a[2], b[2]); 33 | float top = max(a[1], b[1]), bottom = min(a[3], b[3]); 34 | float width = max(right - left + 1, 0.f), height = max(bottom - top + 1, 0.f); 35 | float interS = width * height; 36 | float Sa = (a[2] - a[0] + 1) * (a[3] - a[1] + 1); 37 | float Sb = (b[2] - b[0] + 1) * (b[3] - b[1] + 1); 38 | return interS / (Sa + Sb - interS); 39 | } 40 | 41 | __global__ void nms_kernel(int n_boxes, float nms_overlap_thresh, 42 | float *dev_boxes, unsigned long long *dev_mask) { 43 | const int row_start = blockIdx.y; 44 | const int col_start = blockIdx.x; 45 | 46 | // if (row_start > col_start) return; 47 | 48 | const int row_size = 49 | min(n_boxes - row_start * threadsPerBlock, threadsPerBlock); 50 | const int col_size = 51 | min(n_boxes - col_start * threadsPerBlock, threadsPerBlock); 52 | 53 | __shared__ float block_boxes[threadsPerBlock * 5]; 54 | if (threadIdx.x < col_size) { 55 | block_boxes[threadIdx.x * 5 + 0] = 56 | dev_boxes[(threadsPerBlock * col_start + threadIdx.x) * 5 + 0]; 57 | block_boxes[threadIdx.x * 5 + 1] = 58 | dev_boxes[(threadsPerBlock * col_start + threadIdx.x) * 5 + 1]; 59 | block_boxes[threadIdx.x * 5 + 2] = 60 | dev_boxes[(threadsPerBlock * col_start + threadIdx.x) * 5 + 2]; 61 | block_boxes[threadIdx.x * 5 + 3] = 62 | dev_boxes[(threadsPerBlock * col_start + threadIdx.x) * 5 + 3]; 63 | block_boxes[threadIdx.x * 5 + 4] = 64 | dev_boxes[(threadsPerBlock * col_start + threadIdx.x) * 5 + 4]; 65 | } 66 | __syncthreads(); 67 | 68 | if (threadIdx.x < row_size) { 69 | const int cur_box_idx = threadsPerBlock * row_start + threadIdx.x; 70 | const float *cur_box = dev_boxes + cur_box_idx * 5; 71 | int i = 0; 72 | unsigned long long t = 0; 73 | int start = 0; 74 | if (row_start == col_start) { 75 | start = threadIdx.x + 1; 76 | } 77 | for (i = start; i < col_size; i++) { 78 | if (devIoU(cur_box, block_boxes + i * 5) > nms_overlap_thresh) { 79 | t |= 1ULL << i; 80 | } 81 | } 82 | const int col_blocks = DIVUP(n_boxes, threadsPerBlock); 83 | dev_mask[cur_box_idx * col_blocks + col_start] = t; 84 | } 85 | } 86 | 87 | void nms_cuda_compute(int* keep_out, int *num_out, float* boxes_host, int boxes_num, 88 | int boxes_dim, float nms_overlap_thresh) { 89 | 90 | float* boxes_dev = NULL; 91 | unsigned long long* mask_dev = NULL; 92 | 93 | const int col_blocks = DIVUP(boxes_num, threadsPerBlock); 94 | 95 | CUDA_CHECK(cudaMalloc(&boxes_dev, 96 | boxes_num * boxes_dim * sizeof(float))); 97 | CUDA_CHECK(cudaMemcpy(boxes_dev, 98 | boxes_host, 99 | boxes_num * boxes_dim * sizeof(float), 100 | cudaMemcpyHostToDevice)); 101 | 102 | CUDA_CHECK(cudaMalloc(&mask_dev, 103 | boxes_num * col_blocks * sizeof(unsigned long long))); 104 | 105 | dim3 blocks(DIVUP(boxes_num, threadsPerBlock), 106 | DIVUP(boxes_num, threadsPerBlock)); 107 | dim3 threads(threadsPerBlock); 108 | 109 | // printf("i am at line %d\n", boxes_num); 110 | // printf("i am at line %d\n", boxes_dim); 111 | 112 | nms_kernel<<>>(boxes_num, 113 | nms_overlap_thresh, 114 | boxes_dev, 115 | mask_dev); 116 | 117 | std::vector mask_host(boxes_num * col_blocks); 118 | CUDA_CHECK(cudaMemcpy(&mask_host[0], 119 | mask_dev, 120 | sizeof(unsigned long long) * boxes_num * col_blocks, 121 | cudaMemcpyDeviceToHost)); 122 | 123 | std::vector remv(col_blocks); 124 | memset(&remv[0], 0, sizeof(unsigned long long) * col_blocks); 125 | 126 | // we need to create a memory for keep_out on cpu 127 | // otherwise, the following code cannot run 128 | 129 | int* keep_out_cpu = new int[boxes_num]; 130 | 131 | int num_to_keep = 0; 132 | for (int i = 0; i < boxes_num; i++) { 133 | int nblock = i / threadsPerBlock; 134 | int inblock = i % threadsPerBlock; 135 | 136 | if (!(remv[nblock] & (1ULL << inblock))) { 137 | // orignal: keep_out[num_to_keep++] = i; 138 | keep_out_cpu[num_to_keep++] = i; 139 | unsigned long long *p = &mask_host[0] + i * col_blocks; 140 | for (int j = nblock; j < col_blocks; j++) { 141 | remv[j] |= p[j]; 142 | } 143 | } 144 | } 145 | 146 | // copy keep_out_cpu to keep_out on gpu 147 | CUDA_WARN(cudaMemcpy(keep_out, keep_out_cpu, boxes_num * sizeof(int),cudaMemcpyHostToDevice)); 148 | 149 | // *num_out = num_to_keep; 150 | 151 | // original: *num_out = num_to_keep; 152 | // copy num_to_keep to num_out on gpu 153 | 154 | CUDA_WARN(cudaMemcpy(num_out, &num_to_keep, 1 * sizeof(int),cudaMemcpyHostToDevice)); 155 | 156 | // release cuda memory 157 | CUDA_CHECK(cudaFree(boxes_dev)); 158 | CUDA_CHECK(cudaFree(mask_dev)); 159 | // release cpu memory 160 | delete []keep_out_cpu; 161 | } 162 | -------------------------------------------------------------------------------- /lib/model/nms/src/nms_cuda_kernel.h: -------------------------------------------------------------------------------- 1 | #ifdef __cplusplus 2 | extern "C" { 3 | #endif 4 | 5 | void nms_cuda_compute(int* keep_out, int *num_out, float* boxes_host, int boxes_num, 6 | int boxes_dim, float nms_overlap_thresh); 7 | 8 | #ifdef __cplusplus 9 | } 10 | #endif 11 | -------------------------------------------------------------------------------- /lib/model/roi_align/__init__.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/basiclab/DA-OD-MEAA-PyTorch/ab590ddacdd06886fdbd56f7c4f59cd36167d100/lib/model/roi_align/__init__.py -------------------------------------------------------------------------------- /lib/model/roi_align/_ext/__init__.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/basiclab/DA-OD-MEAA-PyTorch/ab590ddacdd06886fdbd56f7c4f59cd36167d100/lib/model/roi_align/_ext/__init__.py -------------------------------------------------------------------------------- /lib/model/roi_align/_ext/roi_align/__init__.py: -------------------------------------------------------------------------------- 1 | 2 | from torch.utils.ffi import _wrap_function 3 | from ._roi_align import lib as _lib, ffi as _ffi 4 | 5 | __all__ = [] 6 | def _import_symbols(locals): 7 | for symbol in dir(_lib): 8 | fn = getattr(_lib, symbol) 9 | if callable(fn): 10 | locals[symbol] = _wrap_function(fn, _ffi) 11 | else: 12 | locals[symbol] = fn 13 | __all__.append(symbol) 14 | 15 | _import_symbols(locals()) 16 | -------------------------------------------------------------------------------- /lib/model/roi_align/build.py: -------------------------------------------------------------------------------- 1 | from __future__ import print_function 2 | import os 3 | import torch 4 | from torch.utils.ffi import create_extension 5 | 6 | sources = ['src/roi_align.c'] 7 | headers = ['src/roi_align.h'] 8 | extra_objects = [] 9 | #sources = [] 10 | #headers = [] 11 | defines = [] 12 | with_cuda = False 13 | 14 | this_file = os.path.dirname(os.path.realpath(__file__)) 15 | print(this_file) 16 | 17 | if torch.cuda.is_available(): 18 | print('Including CUDA code.') 19 | sources += ['src/roi_align_cuda.c'] 20 | headers += ['src/roi_align_cuda.h'] 21 | defines += [('WITH_CUDA', None)] 22 | with_cuda = True 23 | 24 | extra_objects = ['src/roi_align_kernel.cu.o'] 25 | extra_objects = [os.path.join(this_file, fname) for fname in extra_objects] 26 | 27 | ffi = create_extension( 28 | '_ext.roi_align', 29 | headers=headers, 30 | sources=sources, 31 | define_macros=defines, 32 | relative_to=__file__, 33 | with_cuda=with_cuda, 34 | extra_objects=extra_objects 35 | ) 36 | 37 | if __name__ == '__main__': 38 | ffi.build() 39 | -------------------------------------------------------------------------------- /lib/model/roi_align/functions/__init__.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/basiclab/DA-OD-MEAA-PyTorch/ab590ddacdd06886fdbd56f7c4f59cd36167d100/lib/model/roi_align/functions/__init__.py -------------------------------------------------------------------------------- /lib/model/roi_align/functions/roi_align.py: -------------------------------------------------------------------------------- 1 | import torch 2 | from torch.autograd import Function 3 | from .._ext import roi_align 4 | 5 | 6 | # TODO use save_for_backward instead 7 | class RoIAlignFunction(Function): 8 | def __init__(self, aligned_height, aligned_width, spatial_scale): 9 | self.aligned_width = int(aligned_width) 10 | self.aligned_height = int(aligned_height) 11 | self.spatial_scale = float(spatial_scale) 12 | self.rois = None 13 | self.feature_size = None 14 | 15 | def forward(self, features, rois): 16 | self.rois = rois 17 | self.feature_size = features.size() 18 | 19 | batch_size, num_channels, data_height, data_width = features.size() 20 | num_rois = rois.size(0) 21 | 22 | output = features.new(num_rois, num_channels, self.aligned_height, self.aligned_width).zero_() 23 | if features.is_cuda: 24 | roi_align.roi_align_forward_cuda(self.aligned_height, 25 | self.aligned_width, 26 | self.spatial_scale, features, 27 | rois, output) 28 | else: 29 | roi_align.roi_align_forward(self.aligned_height, 30 | self.aligned_width, 31 | self.spatial_scale, features, 32 | rois, output) 33 | # raise NotImplementedError 34 | 35 | return output 36 | 37 | def backward(self, grad_output): 38 | assert(self.feature_size is not None and grad_output.is_cuda) 39 | 40 | batch_size, num_channels, data_height, data_width = self.feature_size 41 | 42 | grad_input = self.rois.new(batch_size, num_channels, data_height, 43 | data_width).zero_() 44 | roi_align.roi_align_backward_cuda(self.aligned_height, 45 | self.aligned_width, 46 | self.spatial_scale, grad_output, 47 | self.rois, grad_input) 48 | 49 | # print grad_input 50 | 51 | return grad_input, None 52 | -------------------------------------------------------------------------------- /lib/model/roi_align/make.sh: -------------------------------------------------------------------------------- 1 | #!/usr/bin/env bash 2 | 3 | CUDA_PATH=/usr/local/cuda/ 4 | 5 | cd src 6 | echo "Compiling my_lib kernels by nvcc..." 7 | nvcc -c -o roi_align_kernel.cu.o roi_align_kernel.cu -x cu -Xcompiler -fPIC -arch=sm_52 8 | 9 | cd ../ 10 | python build.py 11 | -------------------------------------------------------------------------------- /lib/model/roi_align/modules/__init__.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/basiclab/DA-OD-MEAA-PyTorch/ab590ddacdd06886fdbd56f7c4f59cd36167d100/lib/model/roi_align/modules/__init__.py -------------------------------------------------------------------------------- /lib/model/roi_align/modules/roi_align.py: -------------------------------------------------------------------------------- 1 | from torch.nn.modules.module import Module 2 | from torch.nn.functional import avg_pool2d, max_pool2d 3 | from ..functions.roi_align import RoIAlignFunction 4 | 5 | 6 | class RoIAlign(Module): 7 | def __init__(self, aligned_height, aligned_width, spatial_scale): 8 | super(RoIAlign, self).__init__() 9 | 10 | self.aligned_width = int(aligned_width) 11 | self.aligned_height = int(aligned_height) 12 | self.spatial_scale = float(spatial_scale) 13 | 14 | def forward(self, features, rois): 15 | return RoIAlignFunction(self.aligned_height, self.aligned_width, 16 | self.spatial_scale)(features, rois) 17 | 18 | class RoIAlignAvg(Module): 19 | def __init__(self, aligned_height, aligned_width, spatial_scale): 20 | super(RoIAlignAvg, self).__init__() 21 | 22 | self.aligned_width = int(aligned_width) 23 | self.aligned_height = int(aligned_height) 24 | self.spatial_scale = float(spatial_scale) 25 | 26 | def forward(self, features, rois): 27 | x = RoIAlignFunction(self.aligned_height+1, self.aligned_width+1, 28 | self.spatial_scale)(features, rois) 29 | return avg_pool2d(x, kernel_size=2, stride=1) 30 | 31 | class RoIAlignMax(Module): 32 | def __init__(self, aligned_height, aligned_width, spatial_scale): 33 | super(RoIAlignMax, self).__init__() 34 | 35 | self.aligned_width = int(aligned_width) 36 | self.aligned_height = int(aligned_height) 37 | self.spatial_scale = float(spatial_scale) 38 | 39 | def forward(self, features, rois): 40 | x = RoIAlignFunction(self.aligned_height+1, self.aligned_width+1, 41 | self.spatial_scale)(features, rois) 42 | return max_pool2d(x, kernel_size=2, stride=1) 43 | -------------------------------------------------------------------------------- /lib/model/roi_align/src/roi_align.c: -------------------------------------------------------------------------------- 1 | #include 2 | #include 3 | #include 4 | 5 | 6 | void ROIAlignForwardCpu(const float* bottom_data, const float spatial_scale, const int num_rois, 7 | const int height, const int width, const int channels, 8 | const int aligned_height, const int aligned_width, const float * bottom_rois, 9 | float* top_data); 10 | 11 | void ROIAlignBackwardCpu(const float* top_diff, const float spatial_scale, const int num_rois, 12 | const int height, const int width, const int channels, 13 | const int aligned_height, const int aligned_width, const float * bottom_rois, 14 | float* top_data); 15 | 16 | int roi_align_forward(int aligned_height, int aligned_width, float spatial_scale, 17 | THFloatTensor * features, THFloatTensor * rois, THFloatTensor * output) 18 | { 19 | //Grab the input tensor 20 | float * data_flat = THFloatTensor_data(features); 21 | float * rois_flat = THFloatTensor_data(rois); 22 | 23 | float * output_flat = THFloatTensor_data(output); 24 | 25 | // Number of ROIs 26 | int num_rois = THFloatTensor_size(rois, 0); 27 | int size_rois = THFloatTensor_size(rois, 1); 28 | if (size_rois != 5) 29 | { 30 | return 0; 31 | } 32 | 33 | // data height 34 | int data_height = THFloatTensor_size(features, 2); 35 | // data width 36 | int data_width = THFloatTensor_size(features, 3); 37 | // Number of channels 38 | int num_channels = THFloatTensor_size(features, 1); 39 | 40 | // do ROIAlignForward 41 | ROIAlignForwardCpu(data_flat, spatial_scale, num_rois, data_height, data_width, num_channels, 42 | aligned_height, aligned_width, rois_flat, output_flat); 43 | 44 | return 1; 45 | } 46 | 47 | int roi_align_backward(int aligned_height, int aligned_width, float spatial_scale, 48 | THFloatTensor * top_grad, THFloatTensor * rois, THFloatTensor * bottom_grad) 49 | { 50 | //Grab the input tensor 51 | float * top_grad_flat = THFloatTensor_data(top_grad); 52 | float * rois_flat = THFloatTensor_data(rois); 53 | 54 | float * bottom_grad_flat = THFloatTensor_data(bottom_grad); 55 | 56 | // Number of ROIs 57 | int num_rois = THFloatTensor_size(rois, 0); 58 | int size_rois = THFloatTensor_size(rois, 1); 59 | if (size_rois != 5) 60 | { 61 | return 0; 62 | } 63 | 64 | // batch size 65 | // int batch_size = THFloatTensor_size(bottom_grad, 0); 66 | // data height 67 | int data_height = THFloatTensor_size(bottom_grad, 2); 68 | // data width 69 | int data_width = THFloatTensor_size(bottom_grad, 3); 70 | // Number of channels 71 | int num_channels = THFloatTensor_size(bottom_grad, 1); 72 | 73 | // do ROIAlignBackward 74 | ROIAlignBackwardCpu(top_grad_flat, spatial_scale, num_rois, data_height, 75 | data_width, num_channels, aligned_height, aligned_width, rois_flat, bottom_grad_flat); 76 | 77 | return 1; 78 | } 79 | 80 | void ROIAlignForwardCpu(const float* bottom_data, const float spatial_scale, const int num_rois, 81 | const int height, const int width, const int channels, 82 | const int aligned_height, const int aligned_width, const float * bottom_rois, 83 | float* top_data) 84 | { 85 | const int output_size = num_rois * aligned_height * aligned_width * channels; 86 | 87 | int idx = 0; 88 | for (idx = 0; idx < output_size; ++idx) 89 | { 90 | // (n, c, ph, pw) is an element in the aligned output 91 | int pw = idx % aligned_width; 92 | int ph = (idx / aligned_width) % aligned_height; 93 | int c = (idx / aligned_width / aligned_height) % channels; 94 | int n = idx / aligned_width / aligned_height / channels; 95 | 96 | float roi_batch_ind = bottom_rois[n * 5 + 0]; 97 | float roi_start_w = bottom_rois[n * 5 + 1] * spatial_scale; 98 | float roi_start_h = bottom_rois[n * 5 + 2] * spatial_scale; 99 | float roi_end_w = bottom_rois[n * 5 + 3] * spatial_scale; 100 | float roi_end_h = bottom_rois[n * 5 + 4] * spatial_scale; 101 | 102 | // Force malformed ROI to be 1x1 103 | float roi_width = fmaxf(roi_end_w - roi_start_w + 1., 0.); 104 | float roi_height = fmaxf(roi_end_h - roi_start_h + 1., 0.); 105 | float bin_size_h = roi_height / (aligned_height - 1.); 106 | float bin_size_w = roi_width / (aligned_width - 1.); 107 | 108 | float h = (float)(ph) * bin_size_h + roi_start_h; 109 | float w = (float)(pw) * bin_size_w + roi_start_w; 110 | 111 | int hstart = fminf(floor(h), height - 2); 112 | int wstart = fminf(floor(w), width - 2); 113 | 114 | int img_start = roi_batch_ind * channels * height * width; 115 | 116 | // bilinear interpolation 117 | if (h < 0 || h >= height || w < 0 || w >= width) 118 | { 119 | top_data[idx] = 0.; 120 | } 121 | else 122 | { 123 | float h_ratio = h - (float)(hstart); 124 | float w_ratio = w - (float)(wstart); 125 | int upleft = img_start + (c * height + hstart) * width + wstart; 126 | int upright = upleft + 1; 127 | int downleft = upleft + width; 128 | int downright = downleft + 1; 129 | 130 | top_data[idx] = bottom_data[upleft] * (1. - h_ratio) * (1. - w_ratio) 131 | + bottom_data[upright] * (1. - h_ratio) * w_ratio 132 | + bottom_data[downleft] * h_ratio * (1. - w_ratio) 133 | + bottom_data[downright] * h_ratio * w_ratio; 134 | } 135 | } 136 | } 137 | 138 | void ROIAlignBackwardCpu(const float* top_diff, const float spatial_scale, const int num_rois, 139 | const int height, const int width, const int channels, 140 | const int aligned_height, const int aligned_width, const float * bottom_rois, 141 | float* bottom_diff) 142 | { 143 | const int output_size = num_rois * aligned_height * aligned_width * channels; 144 | 145 | int idx = 0; 146 | for (idx = 0; idx < output_size; ++idx) 147 | { 148 | // (n, c, ph, pw) is an element in the aligned output 149 | int pw = idx % aligned_width; 150 | int ph = (idx / aligned_width) % aligned_height; 151 | int c = (idx / aligned_width / aligned_height) % channels; 152 | int n = idx / aligned_width / aligned_height / channels; 153 | 154 | float roi_batch_ind = bottom_rois[n * 5 + 0]; 155 | float roi_start_w = bottom_rois[n * 5 + 1] * spatial_scale; 156 | float roi_start_h = bottom_rois[n * 5 + 2] * spatial_scale; 157 | float roi_end_w = bottom_rois[n * 5 + 3] * spatial_scale; 158 | float roi_end_h = bottom_rois[n * 5 + 4] * spatial_scale; 159 | 160 | // Force malformed ROI to be 1x1 161 | float roi_width = fmaxf(roi_end_w - roi_start_w + 1., 0.); 162 | float roi_height = fmaxf(roi_end_h - roi_start_h + 1., 0.); 163 | float bin_size_h = roi_height / (aligned_height - 1.); 164 | float bin_size_w = roi_width / (aligned_width - 1.); 165 | 166 | float h = (float)(ph) * bin_size_h + roi_start_h; 167 | float w = (float)(pw) * bin_size_w + roi_start_w; 168 | 169 | int hstart = fminf(floor(h), height - 2); 170 | int wstart = fminf(floor(w), width - 2); 171 | 172 | int img_start = roi_batch_ind * channels * height * width; 173 | 174 | // bilinear interpolation 175 | if (h < 0 || h >= height || w < 0 || w >= width) 176 | { 177 | float h_ratio = h - (float)(hstart); 178 | float w_ratio = w - (float)(wstart); 179 | int upleft = img_start + (c * height + hstart) * width + wstart; 180 | int upright = upleft + 1; 181 | int downleft = upleft + width; 182 | int downright = downleft + 1; 183 | 184 | bottom_diff[upleft] += top_diff[idx] * (1. - h_ratio) * (1. - w_ratio); 185 | bottom_diff[upright] += top_diff[idx] * (1. - h_ratio) * w_ratio; 186 | bottom_diff[downleft] += top_diff[idx] * h_ratio * (1. - w_ratio); 187 | bottom_diff[downright] += top_diff[idx] * h_ratio * w_ratio; 188 | } 189 | } 190 | } 191 | -------------------------------------------------------------------------------- /lib/model/roi_align/src/roi_align.h: -------------------------------------------------------------------------------- 1 | int roi_align_forward(int aligned_height, int aligned_width, float spatial_scale, 2 | THFloatTensor * features, THFloatTensor * rois, THFloatTensor * output); 3 | 4 | int roi_align_backward(int aligned_height, int aligned_width, float spatial_scale, 5 | THFloatTensor * top_grad, THFloatTensor * rois, THFloatTensor * bottom_grad); 6 | -------------------------------------------------------------------------------- /lib/model/roi_align/src/roi_align_cuda.c: -------------------------------------------------------------------------------- 1 | #include 2 | #include 3 | #include "roi_align_kernel.h" 4 | 5 | extern THCState *state; 6 | 7 | int roi_align_forward_cuda(int aligned_height, int aligned_width, float spatial_scale, 8 | THCudaTensor * features, THCudaTensor * rois, THCudaTensor * output) 9 | { 10 | // Grab the input tensor 11 | float * data_flat = THCudaTensor_data(state, features); 12 | float * rois_flat = THCudaTensor_data(state, rois); 13 | 14 | float * output_flat = THCudaTensor_data(state, output); 15 | 16 | // Number of ROIs 17 | int num_rois = THCudaTensor_size(state, rois, 0); 18 | int size_rois = THCudaTensor_size(state, rois, 1); 19 | if (size_rois != 5) 20 | { 21 | return 0; 22 | } 23 | 24 | // data height 25 | int data_height = THCudaTensor_size(state, features, 2); 26 | // data width 27 | int data_width = THCudaTensor_size(state, features, 3); 28 | // Number of channels 29 | int num_channels = THCudaTensor_size(state, features, 1); 30 | 31 | cudaStream_t stream = THCState_getCurrentStream(state); 32 | 33 | ROIAlignForwardLaucher( 34 | data_flat, spatial_scale, num_rois, data_height, 35 | data_width, num_channels, aligned_height, 36 | aligned_width, rois_flat, 37 | output_flat, stream); 38 | 39 | return 1; 40 | } 41 | 42 | int roi_align_backward_cuda(int aligned_height, int aligned_width, float spatial_scale, 43 | THCudaTensor * top_grad, THCudaTensor * rois, THCudaTensor * bottom_grad) 44 | { 45 | // Grab the input tensor 46 | float * top_grad_flat = THCudaTensor_data(state, top_grad); 47 | float * rois_flat = THCudaTensor_data(state, rois); 48 | 49 | float * bottom_grad_flat = THCudaTensor_data(state, bottom_grad); 50 | 51 | // Number of ROIs 52 | int num_rois = THCudaTensor_size(state, rois, 0); 53 | int size_rois = THCudaTensor_size(state, rois, 1); 54 | if (size_rois != 5) 55 | { 56 | return 0; 57 | } 58 | 59 | // batch size 60 | int batch_size = THCudaTensor_size(state, bottom_grad, 0); 61 | // data height 62 | int data_height = THCudaTensor_size(state, bottom_grad, 2); 63 | // data width 64 | int data_width = THCudaTensor_size(state, bottom_grad, 3); 65 | // Number of channels 66 | int num_channels = THCudaTensor_size(state, bottom_grad, 1); 67 | 68 | cudaStream_t stream = THCState_getCurrentStream(state); 69 | ROIAlignBackwardLaucher( 70 | top_grad_flat, spatial_scale, batch_size, num_rois, data_height, 71 | data_width, num_channels, aligned_height, 72 | aligned_width, rois_flat, 73 | bottom_grad_flat, stream); 74 | 75 | return 1; 76 | } 77 | -------------------------------------------------------------------------------- /lib/model/roi_align/src/roi_align_cuda.h: -------------------------------------------------------------------------------- 1 | int roi_align_forward_cuda(int aligned_height, int aligned_width, float spatial_scale, 2 | THCudaTensor * features, THCudaTensor * rois, THCudaTensor * output); 3 | 4 | int roi_align_backward_cuda(int aligned_height, int aligned_width, float spatial_scale, 5 | THCudaTensor * top_grad, THCudaTensor * rois, THCudaTensor * bottom_grad); 6 | -------------------------------------------------------------------------------- /lib/model/roi_align/src/roi_align_kernel.cu: -------------------------------------------------------------------------------- 1 | #ifdef __cplusplus 2 | extern "C" { 3 | #endif 4 | 5 | #include 6 | #include 7 | #include 8 | #include "roi_align_kernel.h" 9 | 10 | #define CUDA_1D_KERNEL_LOOP(i, n) \ 11 | for (int i = blockIdx.x * blockDim.x + threadIdx.x; i < n; \ 12 | i += blockDim.x * gridDim.x) 13 | 14 | 15 | __global__ void ROIAlignForward(const int nthreads, const float* bottom_data, const float spatial_scale, const int height, const int width, 16 | const int channels, const int aligned_height, const int aligned_width, const float* bottom_rois, float* top_data) { 17 | CUDA_1D_KERNEL_LOOP(index, nthreads) { 18 | // (n, c, ph, pw) is an element in the aligned output 19 | // int n = index; 20 | // int pw = n % aligned_width; 21 | // n /= aligned_width; 22 | // int ph = n % aligned_height; 23 | // n /= aligned_height; 24 | // int c = n % channels; 25 | // n /= channels; 26 | 27 | int pw = index % aligned_width; 28 | int ph = (index / aligned_width) % aligned_height; 29 | int c = (index / aligned_width / aligned_height) % channels; 30 | int n = index / aligned_width / aligned_height / channels; 31 | 32 | // bottom_rois += n * 5; 33 | float roi_batch_ind = bottom_rois[n * 5 + 0]; 34 | float roi_start_w = bottom_rois[n * 5 + 1] * spatial_scale; 35 | float roi_start_h = bottom_rois[n * 5 + 2] * spatial_scale; 36 | float roi_end_w = bottom_rois[n * 5 + 3] * spatial_scale; 37 | float roi_end_h = bottom_rois[n * 5 + 4] * spatial_scale; 38 | 39 | // Force malformed ROIs to be 1x1 40 | float roi_width = fmaxf(roi_end_w - roi_start_w + 1., 0.); 41 | float roi_height = fmaxf(roi_end_h - roi_start_h + 1., 0.); 42 | float bin_size_h = roi_height / (aligned_height - 1.); 43 | float bin_size_w = roi_width / (aligned_width - 1.); 44 | 45 | float h = (float)(ph) * bin_size_h + roi_start_h; 46 | float w = (float)(pw) * bin_size_w + roi_start_w; 47 | 48 | int hstart = fminf(floor(h), height - 2); 49 | int wstart = fminf(floor(w), width - 2); 50 | 51 | int img_start = roi_batch_ind * channels * height * width; 52 | 53 | // bilinear interpolation 54 | if (h < 0 || h >= height || w < 0 || w >= width) { 55 | top_data[index] = 0.; 56 | } else { 57 | float h_ratio = h - (float)(hstart); 58 | float w_ratio = w - (float)(wstart); 59 | int upleft = img_start + (c * height + hstart) * width + wstart; 60 | int upright = upleft + 1; 61 | int downleft = upleft + width; 62 | int downright = downleft + 1; 63 | 64 | top_data[index] = bottom_data[upleft] * (1. - h_ratio) * (1. - w_ratio) 65 | + bottom_data[upright] * (1. - h_ratio) * w_ratio 66 | + bottom_data[downleft] * h_ratio * (1. - w_ratio) 67 | + bottom_data[downright] * h_ratio * w_ratio; 68 | } 69 | } 70 | } 71 | 72 | 73 | int ROIAlignForwardLaucher(const float* bottom_data, const float spatial_scale, const int num_rois, const int height, const int width, 74 | const int channels, const int aligned_height, const int aligned_width, const float* bottom_rois, float* top_data, cudaStream_t stream) { 75 | const int kThreadsPerBlock = 1024; 76 | const int output_size = num_rois * aligned_height * aligned_width * channels; 77 | cudaError_t err; 78 | 79 | 80 | ROIAlignForward<<<(output_size + kThreadsPerBlock - 1) / kThreadsPerBlock, kThreadsPerBlock, 0, stream>>>( 81 | output_size, bottom_data, spatial_scale, height, width, channels, 82 | aligned_height, aligned_width, bottom_rois, top_data); 83 | 84 | err = cudaGetLastError(); 85 | if(cudaSuccess != err) { 86 | fprintf( stderr, "cudaCheckError() failed : %s\n", cudaGetErrorString( err ) ); 87 | exit( -1 ); 88 | } 89 | 90 | return 1; 91 | } 92 | 93 | 94 | __global__ void ROIAlignBackward(const int nthreads, const float* top_diff, const float spatial_scale, const int height, const int width, 95 | const int channels, const int aligned_height, const int aligned_width, float* bottom_diff, const float* bottom_rois) { 96 | CUDA_1D_KERNEL_LOOP(index, nthreads) { 97 | 98 | // (n, c, ph, pw) is an element in the aligned output 99 | int pw = index % aligned_width; 100 | int ph = (index / aligned_width) % aligned_height; 101 | int c = (index / aligned_width / aligned_height) % channels; 102 | int n = index / aligned_width / aligned_height / channels; 103 | 104 | float roi_batch_ind = bottom_rois[n * 5 + 0]; 105 | float roi_start_w = bottom_rois[n * 5 + 1] * spatial_scale; 106 | float roi_start_h = bottom_rois[n * 5 + 2] * spatial_scale; 107 | float roi_end_w = bottom_rois[n * 5 + 3] * spatial_scale; 108 | float roi_end_h = bottom_rois[n * 5 + 4] * spatial_scale; 109 | /* int roi_start_w = round(bottom_rois[1] * spatial_scale); */ 110 | /* int roi_start_h = round(bottom_rois[2] * spatial_scale); */ 111 | /* int roi_end_w = round(bottom_rois[3] * spatial_scale); */ 112 | /* int roi_end_h = round(bottom_rois[4] * spatial_scale); */ 113 | 114 | // Force malformed ROIs to be 1x1 115 | float roi_width = fmaxf(roi_end_w - roi_start_w + 1., 0.); 116 | float roi_height = fmaxf(roi_end_h - roi_start_h + 1., 0.); 117 | float bin_size_h = roi_height / (aligned_height - 1.); 118 | float bin_size_w = roi_width / (aligned_width - 1.); 119 | 120 | float h = (float)(ph) * bin_size_h + roi_start_h; 121 | float w = (float)(pw) * bin_size_w + roi_start_w; 122 | 123 | int hstart = fminf(floor(h), height - 2); 124 | int wstart = fminf(floor(w), width - 2); 125 | 126 | int img_start = roi_batch_ind * channels * height * width; 127 | 128 | // bilinear interpolation 129 | if (!(h < 0 || h >= height || w < 0 || w >= width)) { 130 | float h_ratio = h - (float)(hstart); 131 | float w_ratio = w - (float)(wstart); 132 | int upleft = img_start + (c * height + hstart) * width + wstart; 133 | int upright = upleft + 1; 134 | int downleft = upleft + width; 135 | int downright = downleft + 1; 136 | 137 | atomicAdd(bottom_diff + upleft, top_diff[index] * (1. - h_ratio) * (1 - w_ratio)); 138 | atomicAdd(bottom_diff + upright, top_diff[index] * (1. - h_ratio) * w_ratio); 139 | atomicAdd(bottom_diff + downleft, top_diff[index] * h_ratio * (1 - w_ratio)); 140 | atomicAdd(bottom_diff + downright, top_diff[index] * h_ratio * w_ratio); 141 | } 142 | } 143 | } 144 | 145 | int ROIAlignBackwardLaucher(const float* top_diff, const float spatial_scale, const int batch_size, const int num_rois, const int height, const int width, 146 | const int channels, const int aligned_height, const int aligned_width, const float* bottom_rois, float* bottom_diff, cudaStream_t stream) { 147 | const int kThreadsPerBlock = 1024; 148 | const int output_size = num_rois * aligned_height * aligned_width * channels; 149 | cudaError_t err; 150 | 151 | ROIAlignBackward<<<(output_size + kThreadsPerBlock - 1) / kThreadsPerBlock, kThreadsPerBlock, 0, stream>>>( 152 | output_size, top_diff, spatial_scale, height, width, channels, 153 | aligned_height, aligned_width, bottom_diff, bottom_rois); 154 | 155 | err = cudaGetLastError(); 156 | if(cudaSuccess != err) { 157 | fprintf( stderr, "cudaCheckError() failed : %s\n", cudaGetErrorString( err ) ); 158 | exit( -1 ); 159 | } 160 | 161 | return 1; 162 | } 163 | 164 | 165 | #ifdef __cplusplus 166 | } 167 | #endif 168 | -------------------------------------------------------------------------------- /lib/model/roi_align/src/roi_align_kernel.h: -------------------------------------------------------------------------------- 1 | #ifndef _ROI_ALIGN_KERNEL 2 | #define _ROI_ALIGN_KERNEL 3 | 4 | #ifdef __cplusplus 5 | extern "C" { 6 | #endif 7 | 8 | __global__ void ROIAlignForward(const int nthreads, const float* bottom_data, 9 | const float spatial_scale, const int height, const int width, 10 | const int channels, const int aligned_height, const int aligned_width, 11 | const float* bottom_rois, float* top_data); 12 | 13 | int ROIAlignForwardLaucher( 14 | const float* bottom_data, const float spatial_scale, const int num_rois, const int height, 15 | const int width, const int channels, const int aligned_height, 16 | const int aligned_width, const float* bottom_rois, 17 | float* top_data, cudaStream_t stream); 18 | 19 | __global__ void ROIAlignBackward(const int nthreads, const float* top_diff, 20 | const float spatial_scale, const int height, const int width, 21 | const int channels, const int aligned_height, const int aligned_width, 22 | float* bottom_diff, const float* bottom_rois); 23 | 24 | int ROIAlignBackwardLaucher(const float* top_diff, const float spatial_scale, const int batch_size, const int num_rois, 25 | const int height, const int width, const int channels, const int aligned_height, 26 | const int aligned_width, const float* bottom_rois, 27 | float* bottom_diff, cudaStream_t stream); 28 | 29 | #ifdef __cplusplus 30 | } 31 | #endif 32 | 33 | #endif 34 | 35 | -------------------------------------------------------------------------------- /lib/model/roi_crop/__init__.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/basiclab/DA-OD-MEAA-PyTorch/ab590ddacdd06886fdbd56f7c4f59cd36167d100/lib/model/roi_crop/__init__.py -------------------------------------------------------------------------------- /lib/model/roi_crop/_ext/__init__.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/basiclab/DA-OD-MEAA-PyTorch/ab590ddacdd06886fdbd56f7c4f59cd36167d100/lib/model/roi_crop/_ext/__init__.py -------------------------------------------------------------------------------- /lib/model/roi_crop/_ext/crop_resize/__init__.py: -------------------------------------------------------------------------------- 1 | 2 | from torch.utils.ffi import _wrap_function 3 | from ._crop_resize import lib as _lib, ffi as _ffi 4 | 5 | __all__ = [] 6 | def _import_symbols(locals): 7 | for symbol in dir(_lib): 8 | fn = getattr(_lib, symbol) 9 | locals[symbol] = _wrap_function(fn, _ffi) 10 | __all__.append(symbol) 11 | 12 | _import_symbols(locals()) 13 | -------------------------------------------------------------------------------- /lib/model/roi_crop/_ext/roi_crop/__init__.py: -------------------------------------------------------------------------------- 1 | 2 | from torch.utils.ffi import _wrap_function 3 | from ._roi_crop import lib as _lib, ffi as _ffi 4 | 5 | __all__ = [] 6 | def _import_symbols(locals): 7 | for symbol in dir(_lib): 8 | fn = getattr(_lib, symbol) 9 | if callable(fn): 10 | locals[symbol] = _wrap_function(fn, _ffi) 11 | else: 12 | locals[symbol] = fn 13 | __all__.append(symbol) 14 | 15 | _import_symbols(locals()) 16 | -------------------------------------------------------------------------------- /lib/model/roi_crop/build.py: -------------------------------------------------------------------------------- 1 | from __future__ import print_function 2 | import os 3 | import torch 4 | from torch.utils.ffi import create_extension 5 | 6 | #this_file = os.path.dirname(__file__) 7 | 8 | sources = ['src/roi_crop.c'] 9 | headers = ['src/roi_crop.h'] 10 | defines = [] 11 | with_cuda = False 12 | 13 | if torch.cuda.is_available(): 14 | print('Including CUDA code.') 15 | sources += ['src/roi_crop_cuda.c'] 16 | headers += ['src/roi_crop_cuda.h'] 17 | defines += [('WITH_CUDA', None)] 18 | with_cuda = True 19 | 20 | this_file = os.path.dirname(os.path.realpath(__file__)) 21 | print(this_file) 22 | extra_objects = ['src/roi_crop_cuda_kernel.cu.o'] 23 | extra_objects = [os.path.join(this_file, fname) for fname in extra_objects] 24 | 25 | ffi = create_extension( 26 | '_ext.roi_crop', 27 | headers=headers, 28 | sources=sources, 29 | define_macros=defines, 30 | relative_to=__file__, 31 | with_cuda=with_cuda, 32 | extra_objects=extra_objects 33 | ) 34 | 35 | if __name__ == '__main__': 36 | ffi.build() 37 | -------------------------------------------------------------------------------- /lib/model/roi_crop/functions/__init__.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/basiclab/DA-OD-MEAA-PyTorch/ab590ddacdd06886fdbd56f7c4f59cd36167d100/lib/model/roi_crop/functions/__init__.py -------------------------------------------------------------------------------- /lib/model/roi_crop/functions/crop_resize.py: -------------------------------------------------------------------------------- 1 | # functions/add.py 2 | import torch 3 | from torch.autograd import Function 4 | from .._ext import roi_crop 5 | from cffi import FFI 6 | ffi = FFI() 7 | 8 | class RoICropFunction(Function): 9 | def forward(self, input1, input2): 10 | self.input1 = input1 11 | self.input2 = input2 12 | self.device_c = ffi.new("int *") 13 | output = torch.zeros(input2.size()[0], input1.size()[1], input2.size()[1], input2.size()[2]) 14 | #print('decice %d' % torch.cuda.current_device()) 15 | if input1.is_cuda: 16 | self.device = torch.cuda.current_device() 17 | else: 18 | self.device = -1 19 | self.device_c[0] = self.device 20 | if not input1.is_cuda: 21 | roi_crop.BilinearSamplerBHWD_updateOutput(input1, input2, output) 22 | else: 23 | output = output.cuda(self.device) 24 | roi_crop.BilinearSamplerBHWD_updateOutput_cuda(input1, input2, output) 25 | return output 26 | 27 | def backward(self, grad_output): 28 | grad_input1 = torch.zeros(self.input1.size()) 29 | grad_input2 = torch.zeros(self.input2.size()) 30 | #print('backward decice %d' % self.device) 31 | if not grad_output.is_cuda: 32 | roi_crop.BilinearSamplerBHWD_updateGradInput(self.input1, self.input2, grad_input1, grad_input2, grad_output) 33 | else: 34 | grad_input1 = grad_input1.cuda(self.device) 35 | grad_input2 = grad_input2.cuda(self.device) 36 | roi_crop.BilinearSamplerBHWD_updateGradInput_cuda(self.input1, self.input2, grad_input1, grad_input2, grad_output) 37 | return grad_input1, grad_input2 38 | -------------------------------------------------------------------------------- /lib/model/roi_crop/functions/gridgen.py: -------------------------------------------------------------------------------- 1 | # functions/add.py 2 | import torch 3 | from torch.autograd import Function 4 | import numpy as np 5 | 6 | 7 | class AffineGridGenFunction(Function): 8 | def __init__(self, height, width,lr=1): 9 | super(AffineGridGenFunction, self).__init__() 10 | self.lr = lr 11 | self.height, self.width = height, width 12 | self.grid = np.zeros( [self.height, self.width, 3], dtype=np.float32) 13 | self.grid[:,:,0] = np.expand_dims(np.repeat(np.expand_dims(np.arange(-1, 1, 2.0/(self.height)), 0), repeats = self.width, axis = 0).T, 0) 14 | self.grid[:,:,1] = np.expand_dims(np.repeat(np.expand_dims(np.arange(-1, 1, 2.0/(self.width)), 0), repeats = self.height, axis = 0), 0) 15 | # self.grid[:,:,0] = np.expand_dims(np.repeat(np.expand_dims(np.arange(-1, 1, 2.0/(self.height - 1)), 0), repeats = self.width, axis = 0).T, 0) 16 | # self.grid[:,:,1] = np.expand_dims(np.repeat(np.expand_dims(np.arange(-1, 1, 2.0/(self.width - 1)), 0), repeats = self.height, axis = 0), 0) 17 | self.grid[:,:,2] = np.ones([self.height, width]) 18 | self.grid = torch.from_numpy(self.grid.astype(np.float32)) 19 | #print(self.grid) 20 | 21 | def forward(self, input1): 22 | self.input1 = input1 23 | output = input1.new(torch.Size([input1.size(0)]) + self.grid.size()).zero_() 24 | self.batchgrid = input1.new(torch.Size([input1.size(0)]) + self.grid.size()).zero_() 25 | for i in range(input1.size(0)): 26 | self.batchgrid[i] = self.grid.astype(self.batchgrid[i]) 27 | 28 | # if input1.is_cuda: 29 | # self.batchgrid = self.batchgrid.cuda() 30 | # output = output.cuda() 31 | 32 | for i in range(input1.size(0)): 33 | output = torch.bmm(self.batchgrid.view(-1, self.height*self.width, 3), torch.transpose(input1, 1, 2)).view(-1, self.height, self.width, 2) 34 | 35 | return output 36 | 37 | def backward(self, grad_output): 38 | 39 | grad_input1 = self.input1.new(self.input1.size()).zero_() 40 | 41 | # if grad_output.is_cuda: 42 | # self.batchgrid = self.batchgrid.cuda() 43 | # grad_input1 = grad_input1.cuda() 44 | 45 | grad_input1 = torch.baddbmm(grad_input1, torch.transpose(grad_output.view(-1, self.height*self.width, 2), 1,2), self.batchgrid.view(-1, self.height*self.width, 3)) 46 | return grad_input1 47 | -------------------------------------------------------------------------------- /lib/model/roi_crop/functions/roi_crop.py: -------------------------------------------------------------------------------- 1 | # functions/add.py 2 | import torch 3 | from torch.autograd import Function 4 | from .._ext import roi_crop 5 | import pdb 6 | 7 | class RoICropFunction(Function): 8 | def forward(self, input1, input2): 9 | self.input1 = input1.clone() 10 | self.input2 = input2.clone() 11 | output = input2.new(input2.size()[0], input1.size()[1], input2.size()[1], input2.size()[2]).zero_() 12 | assert output.get_device() == input1.get_device(), "output and input1 must on the same device" 13 | assert output.get_device() == input2.get_device(), "output and input2 must on the same device" 14 | roi_crop.BilinearSamplerBHWD_updateOutput_cuda(input1, input2, output) 15 | return output 16 | 17 | def backward(self, grad_output): 18 | grad_input1 = self.input1.new(self.input1.size()).zero_() 19 | grad_input2 = self.input2.new(self.input2.size()).zero_() 20 | roi_crop.BilinearSamplerBHWD_updateGradInput_cuda(self.input1, self.input2, grad_input1, grad_input2, grad_output) 21 | return grad_input1, grad_input2 22 | -------------------------------------------------------------------------------- /lib/model/roi_crop/make.sh: -------------------------------------------------------------------------------- 1 | #!/usr/bin/env bash 2 | 3 | CUDA_PATH=/usr/local/cuda/ 4 | 5 | cd src 6 | echo "Compiling my_lib kernels by nvcc..." 7 | nvcc -c -o roi_crop_cuda_kernel.cu.o roi_crop_cuda_kernel.cu -x cu -Xcompiler -fPIC -arch=sm_52 8 | 9 | cd ../ 10 | python build.py 11 | -------------------------------------------------------------------------------- /lib/model/roi_crop/modules/__init__.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/basiclab/DA-OD-MEAA-PyTorch/ab590ddacdd06886fdbd56f7c4f59cd36167d100/lib/model/roi_crop/modules/__init__.py -------------------------------------------------------------------------------- /lib/model/roi_crop/modules/roi_crop.py: -------------------------------------------------------------------------------- 1 | from torch.nn.modules.module import Module 2 | from ..functions.roi_crop import RoICropFunction 3 | 4 | class _RoICrop(Module): 5 | def __init__(self, layout = 'BHWD'): 6 | super(_RoICrop, self).__init__() 7 | def forward(self, input1, input2): 8 | return RoICropFunction()(input1, input2) 9 | -------------------------------------------------------------------------------- /lib/model/roi_crop/src/roi_crop.h: -------------------------------------------------------------------------------- 1 | int BilinearSamplerBHWD_updateOutput(THFloatTensor *inputImages, THFloatTensor *grids, THFloatTensor *output); 2 | 3 | int BilinearSamplerBHWD_updateGradInput(THFloatTensor *inputImages, THFloatTensor *grids, THFloatTensor *gradInputImages, 4 | THFloatTensor *gradGrids, THFloatTensor *gradOutput); 5 | 6 | 7 | 8 | int BilinearSamplerBCHW_updateOutput(THFloatTensor *inputImages, THFloatTensor *grids, THFloatTensor *output); 9 | 10 | int BilinearSamplerBCHW_updateGradInput(THFloatTensor *inputImages, THFloatTensor *grids, THFloatTensor *gradInputImages, 11 | THFloatTensor *gradGrids, THFloatTensor *gradOutput); 12 | -------------------------------------------------------------------------------- /lib/model/roi_crop/src/roi_crop_cuda.c: -------------------------------------------------------------------------------- 1 | #include 2 | #include 3 | #include 4 | #include "roi_crop_cuda_kernel.h" 5 | 6 | #define real float 7 | 8 | // this symbol will be resolved automatically from PyTorch libs 9 | extern THCState *state; 10 | 11 | // Bilinear sampling is done in BHWD (coalescing is not obvious in BDHW) 12 | // we assume BHWD format in inputImages 13 | // we assume BHW(YX) format on grids 14 | 15 | int BilinearSamplerBHWD_updateOutput_cuda(THCudaTensor *inputImages, THCudaTensor *grids, THCudaTensor *output){ 16 | // THCState *state = getCutorchState(L); 17 | // THCudaTensor *inputImages = (THCudaTensor *)luaT_checkudata(L, 2, "torch.CudaTensor"); 18 | // THCudaTensor *grids = (THCudaTensor *)luaT_checkudata(L, 3, "torch.CudaTensor"); 19 | // THCudaTensor *output = (THCudaTensor *)luaT_checkudata(L, 4, "torch.CudaTensor"); 20 | 21 | int success = 0; 22 | success = BilinearSamplerBHWD_updateOutput_cuda_kernel(THCudaTensor_size(state, output, 1), 23 | THCudaTensor_size(state, output, 3), 24 | THCudaTensor_size(state, output, 2), 25 | THCudaTensor_size(state, output, 0), 26 | THCudaTensor_size(state, inputImages, 1), 27 | THCudaTensor_size(state, inputImages, 2), 28 | THCudaTensor_size(state, inputImages, 3), 29 | THCudaTensor_size(state, inputImages, 0), 30 | THCudaTensor_data(state, inputImages), 31 | THCudaTensor_stride(state, inputImages, 0), 32 | THCudaTensor_stride(state, inputImages, 1), 33 | THCudaTensor_stride(state, inputImages, 2), 34 | THCudaTensor_stride(state, inputImages, 3), 35 | THCudaTensor_data(state, grids), 36 | THCudaTensor_stride(state, grids, 0), 37 | THCudaTensor_stride(state, grids, 3), 38 | THCudaTensor_stride(state, grids, 1), 39 | THCudaTensor_stride(state, grids, 2), 40 | THCudaTensor_data(state, output), 41 | THCudaTensor_stride(state, output, 0), 42 | THCudaTensor_stride(state, output, 1), 43 | THCudaTensor_stride(state, output, 2), 44 | THCudaTensor_stride(state, output, 3), 45 | THCState_getCurrentStream(state)); 46 | 47 | //check for errors 48 | if (!success) { 49 | THError("aborting"); 50 | } 51 | return 1; 52 | } 53 | 54 | int BilinearSamplerBHWD_updateGradInput_cuda(THCudaTensor *inputImages, THCudaTensor *grids, THCudaTensor *gradInputImages, 55 | THCudaTensor *gradGrids, THCudaTensor *gradOutput) 56 | { 57 | // THCState *state = getCutorchState(L); 58 | // THCudaTensor *inputImages = (THCudaTensor *)luaT_checkudata(L, 2, "torch.CudaTensor"); 59 | // THCudaTensor *grids = (THCudaTensor *)luaT_checkudata(L, 3, "torch.CudaTensor"); 60 | // THCudaTensor *gradInputImages = (THCudaTensor *)luaT_checkudata(L, 4, "torch.CudaTensor"); 61 | // THCudaTensor *gradGrids = (THCudaTensor *)luaT_checkudata(L, 5, "torch.CudaTensor"); 62 | // THCudaTensor *gradOutput = (THCudaTensor *)luaT_checkudata(L, 6, "torch.CudaTensor"); 63 | 64 | int success = 0; 65 | success = BilinearSamplerBHWD_updateGradInput_cuda_kernel(THCudaTensor_size(state, gradOutput, 1), 66 | THCudaTensor_size(state, gradOutput, 3), 67 | THCudaTensor_size(state, gradOutput, 2), 68 | THCudaTensor_size(state, gradOutput, 0), 69 | THCudaTensor_size(state, inputImages, 1), 70 | THCudaTensor_size(state, inputImages, 2), 71 | THCudaTensor_size(state, inputImages, 3), 72 | THCudaTensor_size(state, inputImages, 0), 73 | THCudaTensor_data(state, inputImages), 74 | THCudaTensor_stride(state, inputImages, 0), 75 | THCudaTensor_stride(state, inputImages, 1), 76 | THCudaTensor_stride(state, inputImages, 2), 77 | THCudaTensor_stride(state, inputImages, 3), 78 | THCudaTensor_data(state, grids), 79 | THCudaTensor_stride(state, grids, 0), 80 | THCudaTensor_stride(state, grids, 3), 81 | THCudaTensor_stride(state, grids, 1), 82 | THCudaTensor_stride(state, grids, 2), 83 | THCudaTensor_data(state, gradInputImages), 84 | THCudaTensor_stride(state, gradInputImages, 0), 85 | THCudaTensor_stride(state, gradInputImages, 1), 86 | THCudaTensor_stride(state, gradInputImages, 2), 87 | THCudaTensor_stride(state, gradInputImages, 3), 88 | THCudaTensor_data(state, gradGrids), 89 | THCudaTensor_stride(state, gradGrids, 0), 90 | THCudaTensor_stride(state, gradGrids, 3), 91 | THCudaTensor_stride(state, gradGrids, 1), 92 | THCudaTensor_stride(state, gradGrids, 2), 93 | THCudaTensor_data(state, gradOutput), 94 | THCudaTensor_stride(state, gradOutput, 0), 95 | THCudaTensor_stride(state, gradOutput, 1), 96 | THCudaTensor_stride(state, gradOutput, 2), 97 | THCudaTensor_stride(state, gradOutput, 3), 98 | THCState_getCurrentStream(state)); 99 | 100 | //check for errors 101 | if (!success) { 102 | THError("aborting"); 103 | } 104 | return 1; 105 | } 106 | -------------------------------------------------------------------------------- /lib/model/roi_crop/src/roi_crop_cuda.h: -------------------------------------------------------------------------------- 1 | // Bilinear sampling is done in BHWD (coalescing is not obvious in BDHW) 2 | // we assume BHWD format in inputImages 3 | // we assume BHW(YX) format on grids 4 | 5 | int BilinearSamplerBHWD_updateOutput_cuda(THCudaTensor *inputImages, THCudaTensor *grids, THCudaTensor *output); 6 | 7 | int BilinearSamplerBHWD_updateGradInput_cuda(THCudaTensor *inputImages, THCudaTensor *grids, THCudaTensor *gradInputImages, 8 | THCudaTensor *gradGrids, THCudaTensor *gradOutput); 9 | -------------------------------------------------------------------------------- /lib/model/roi_crop/src/roi_crop_cuda_kernel.h: -------------------------------------------------------------------------------- 1 | #ifdef __cplusplus 2 | extern "C" { 3 | #endif 4 | 5 | 6 | int BilinearSamplerBHWD_updateOutput_cuda_kernel(/*output->size[3]*/int oc, 7 | /*output->size[2]*/int ow, 8 | /*output->size[1]*/int oh, 9 | /*output->size[0]*/int ob, 10 | /*THCudaTensor_size(state, inputImages, 3)*/int ic, 11 | /*THCudaTensor_size(state, inputImages, 1)*/int ih, 12 | /*THCudaTensor_size(state, inputImages, 2)*/int iw, 13 | /*THCudaTensor_size(state, inputImages, 0)*/int ib, 14 | /*THCudaTensor *inputImages*/float *inputImages, int isb, int isc, int ish, int isw, 15 | /*THCudaTensor *grids*/float *grids, int gsb, int gsc, int gsh, int gsw, 16 | /*THCudaTensor *output*/float *output, int osb, int osc, int osh, int osw, 17 | /*THCState_getCurrentStream(state)*/cudaStream_t stream); 18 | 19 | int BilinearSamplerBHWD_updateGradInput_cuda_kernel(/*gradOutput->size[3]*/int goc, 20 | /*gradOutput->size[2]*/int gow, 21 | /*gradOutput->size[1]*/int goh, 22 | /*gradOutput->size[0]*/int gob, 23 | /*THCudaTensor_size(state, inputImages, 3)*/int ic, 24 | /*THCudaTensor_size(state, inputImages, 1)*/int ih, 25 | /*THCudaTensor_size(state, inputImages, 2)*/int iw, 26 | /*THCudaTensor_size(state, inputImages, 0)*/int ib, 27 | /*THCudaTensor *inputImages*/float *inputImages, int isb, int isc, int ish, int isw, 28 | /*THCudaTensor *grids*/float *grids, int gsb, int gsc, int gsh, int gsw, 29 | /*THCudaTensor *gradInputImages*/float *gradInputImages, int gisb, int gisc, int gish, int gisw, 30 | /*THCudaTensor *gradGrids*/float *gradGrids, int ggsb, int ggsc, int ggsh, int ggsw, 31 | /*THCudaTensor *gradOutput*/float *gradOutput, int gosb, int gosc, int gosh, int gosw, 32 | /*THCState_getCurrentStream(state)*/cudaStream_t stream); 33 | 34 | 35 | #ifdef __cplusplus 36 | } 37 | #endif 38 | -------------------------------------------------------------------------------- /lib/model/roi_pooling/__init__.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/basiclab/DA-OD-MEAA-PyTorch/ab590ddacdd06886fdbd56f7c4f59cd36167d100/lib/model/roi_pooling/__init__.py -------------------------------------------------------------------------------- /lib/model/roi_pooling/_ext/__init__.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/basiclab/DA-OD-MEAA-PyTorch/ab590ddacdd06886fdbd56f7c4f59cd36167d100/lib/model/roi_pooling/_ext/__init__.py -------------------------------------------------------------------------------- /lib/model/roi_pooling/_ext/roi_pooling/__init__.py: -------------------------------------------------------------------------------- 1 | 2 | from torch.utils.ffi import _wrap_function 3 | from ._roi_pooling import lib as _lib, ffi as _ffi 4 | 5 | __all__ = [] 6 | def _import_symbols(locals): 7 | for symbol in dir(_lib): 8 | fn = getattr(_lib, symbol) 9 | if callable(fn): 10 | locals[symbol] = _wrap_function(fn, _ffi) 11 | else: 12 | locals[symbol] = fn 13 | __all__.append(symbol) 14 | 15 | _import_symbols(locals()) 16 | -------------------------------------------------------------------------------- /lib/model/roi_pooling/build.py: -------------------------------------------------------------------------------- 1 | from __future__ import print_function 2 | import os 3 | import torch 4 | from torch.utils.ffi import create_extension 5 | 6 | 7 | sources = ['src/roi_pooling.c'] 8 | headers = ['src/roi_pooling.h'] 9 | extra_objects = [] 10 | defines = [] 11 | with_cuda = False 12 | 13 | this_file = os.path.dirname(os.path.realpath(__file__)) 14 | print(this_file) 15 | 16 | if torch.cuda.is_available(): 17 | print('Including CUDA code.') 18 | sources += ['src/roi_pooling_cuda.c'] 19 | headers += ['src/roi_pooling_cuda.h'] 20 | defines += [('WITH_CUDA', None)] 21 | with_cuda = True 22 | extra_objects = ['src/roi_pooling.cu.o'] 23 | extra_objects = [os.path.join(this_file, fname) for fname in extra_objects] 24 | 25 | ffi = create_extension( 26 | '_ext.roi_pooling', 27 | headers=headers, 28 | sources=sources, 29 | define_macros=defines, 30 | relative_to=__file__, 31 | with_cuda=with_cuda, 32 | extra_objects=extra_objects 33 | ) 34 | 35 | if __name__ == '__main__': 36 | ffi.build() 37 | -------------------------------------------------------------------------------- /lib/model/roi_pooling/functions/__init__.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/basiclab/DA-OD-MEAA-PyTorch/ab590ddacdd06886fdbd56f7c4f59cd36167d100/lib/model/roi_pooling/functions/__init__.py -------------------------------------------------------------------------------- /lib/model/roi_pooling/functions/roi_pool.py: -------------------------------------------------------------------------------- 1 | import torch 2 | from torch.autograd import Function 3 | from .._ext import roi_pooling 4 | import pdb 5 | 6 | class RoIPoolFunction(Function): 7 | def __init__(ctx, pooled_height, pooled_width, spatial_scale): 8 | ctx.pooled_width = pooled_width 9 | ctx.pooled_height = pooled_height 10 | ctx.spatial_scale = spatial_scale 11 | ctx.feature_size = None 12 | 13 | def forward(ctx, features, rois): 14 | ctx.feature_size = features.size() 15 | batch_size, num_channels, data_height, data_width = ctx.feature_size 16 | num_rois = rois.size(0) 17 | output = features.new(num_rois, num_channels, ctx.pooled_height, ctx.pooled_width).zero_() 18 | ctx.argmax = features.new(num_rois, num_channels, ctx.pooled_height, ctx.pooled_width).zero_().int() 19 | ctx.rois = rois 20 | if not features.is_cuda: 21 | _features = features.permute(0, 2, 3, 1) 22 | roi_pooling.roi_pooling_forward(ctx.pooled_height, ctx.pooled_width, ctx.spatial_scale, 23 | _features, rois, output) 24 | else: 25 | roi_pooling.roi_pooling_forward_cuda(ctx.pooled_height, ctx.pooled_width, ctx.spatial_scale, 26 | features, rois, output, ctx.argmax) 27 | 28 | return output 29 | 30 | def backward(ctx, grad_output): 31 | assert(ctx.feature_size is not None and grad_output.is_cuda) 32 | batch_size, num_channels, data_height, data_width = ctx.feature_size 33 | grad_input = grad_output.new(batch_size, num_channels, data_height, data_width).zero_() 34 | 35 | roi_pooling.roi_pooling_backward_cuda(ctx.pooled_height, ctx.pooled_width, ctx.spatial_scale, 36 | grad_output, ctx.rois, grad_input, ctx.argmax) 37 | 38 | return grad_input, None 39 | -------------------------------------------------------------------------------- /lib/model/roi_pooling/modules/__init__.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/basiclab/DA-OD-MEAA-PyTorch/ab590ddacdd06886fdbd56f7c4f59cd36167d100/lib/model/roi_pooling/modules/__init__.py -------------------------------------------------------------------------------- /lib/model/roi_pooling/modules/roi_pool.py: -------------------------------------------------------------------------------- 1 | from torch.nn.modules.module import Module 2 | from ..functions.roi_pool import RoIPoolFunction 3 | 4 | 5 | class _RoIPooling(Module): 6 | def __init__(self, pooled_height, pooled_width, spatial_scale): 7 | super(_RoIPooling, self).__init__() 8 | 9 | self.pooled_width = int(pooled_width) 10 | self.pooled_height = int(pooled_height) 11 | self.spatial_scale = float(spatial_scale) 12 | 13 | def forward(self, features, rois): 14 | return RoIPoolFunction(self.pooled_height, self.pooled_width, self.spatial_scale)(features, rois) 15 | -------------------------------------------------------------------------------- /lib/model/roi_pooling/src/roi_pooling.c: -------------------------------------------------------------------------------- 1 | #include 2 | #include 3 | 4 | int roi_pooling_forward(int pooled_height, int pooled_width, float spatial_scale, 5 | THFloatTensor * features, THFloatTensor * rois, THFloatTensor * output) 6 | { 7 | // Grab the input tensor 8 | float * data_flat = THFloatTensor_data(features); 9 | float * rois_flat = THFloatTensor_data(rois); 10 | 11 | float * output_flat = THFloatTensor_data(output); 12 | 13 | // Number of ROIs 14 | int num_rois = THFloatTensor_size(rois, 0); 15 | int size_rois = THFloatTensor_size(rois, 1); 16 | // batch size 17 | int batch_size = THFloatTensor_size(features, 0); 18 | if(batch_size != 1) 19 | { 20 | return 0; 21 | } 22 | // data height 23 | int data_height = THFloatTensor_size(features, 1); 24 | // data width 25 | int data_width = THFloatTensor_size(features, 2); 26 | // Number of channels 27 | int num_channels = THFloatTensor_size(features, 3); 28 | 29 | // Set all element of the output tensor to -inf. 30 | THFloatStorage_fill(THFloatTensor_storage(output), -1); 31 | 32 | // For each ROI R = [batch_index x1 y1 x2 y2]: max pool over R 33 | int index_roi = 0; 34 | int index_output = 0; 35 | int n; 36 | for (n = 0; n < num_rois; ++n) 37 | { 38 | int roi_batch_ind = rois_flat[index_roi + 0]; 39 | int roi_start_w = round(rois_flat[index_roi + 1] * spatial_scale); 40 | int roi_start_h = round(rois_flat[index_roi + 2] * spatial_scale); 41 | int roi_end_w = round(rois_flat[index_roi + 3] * spatial_scale); 42 | int roi_end_h = round(rois_flat[index_roi + 4] * spatial_scale); 43 | // CHECK_GE(roi_batch_ind, 0); 44 | // CHECK_LT(roi_batch_ind, batch_size); 45 | 46 | int roi_height = fmaxf(roi_end_h - roi_start_h + 1, 1); 47 | int roi_width = fmaxf(roi_end_w - roi_start_w + 1, 1); 48 | float bin_size_h = (float)(roi_height) / (float)(pooled_height); 49 | float bin_size_w = (float)(roi_width) / (float)(pooled_width); 50 | 51 | int index_data = roi_batch_ind * data_height * data_width * num_channels; 52 | const int output_area = pooled_width * pooled_height; 53 | 54 | int c, ph, pw; 55 | for (ph = 0; ph < pooled_height; ++ph) 56 | { 57 | for (pw = 0; pw < pooled_width; ++pw) 58 | { 59 | int hstart = (floor((float)(ph) * bin_size_h)); 60 | int wstart = (floor((float)(pw) * bin_size_w)); 61 | int hend = (ceil((float)(ph + 1) * bin_size_h)); 62 | int wend = (ceil((float)(pw + 1) * bin_size_w)); 63 | 64 | hstart = fminf(fmaxf(hstart + roi_start_h, 0), data_height); 65 | hend = fminf(fmaxf(hend + roi_start_h, 0), data_height); 66 | wstart = fminf(fmaxf(wstart + roi_start_w, 0), data_width); 67 | wend = fminf(fmaxf(wend + roi_start_w, 0), data_width); 68 | 69 | const int pool_index = index_output + (ph * pooled_width + pw); 70 | int is_empty = (hend <= hstart) || (wend <= wstart); 71 | if (is_empty) 72 | { 73 | for (c = 0; c < num_channels * output_area; c += output_area) 74 | { 75 | output_flat[pool_index + c] = 0; 76 | } 77 | } 78 | else 79 | { 80 | int h, w, c; 81 | for (h = hstart; h < hend; ++h) 82 | { 83 | for (w = wstart; w < wend; ++w) 84 | { 85 | for (c = 0; c < num_channels; ++c) 86 | { 87 | const int index = (h * data_width + w) * num_channels + c; 88 | if (data_flat[index_data + index] > output_flat[pool_index + c * output_area]) 89 | { 90 | output_flat[pool_index + c * output_area] = data_flat[index_data + index]; 91 | } 92 | } 93 | } 94 | } 95 | } 96 | } 97 | } 98 | 99 | // Increment ROI index 100 | index_roi += size_rois; 101 | index_output += pooled_height * pooled_width * num_channels; 102 | } 103 | return 1; 104 | } -------------------------------------------------------------------------------- /lib/model/roi_pooling/src/roi_pooling.h: -------------------------------------------------------------------------------- 1 | int roi_pooling_forward(int pooled_height, int pooled_width, float spatial_scale, 2 | THFloatTensor * features, THFloatTensor * rois, THFloatTensor * output); -------------------------------------------------------------------------------- /lib/model/roi_pooling/src/roi_pooling_cuda.c: -------------------------------------------------------------------------------- 1 | #include 2 | #include 3 | #include "roi_pooling_kernel.h" 4 | 5 | extern THCState *state; 6 | 7 | int roi_pooling_forward_cuda(int pooled_height, int pooled_width, float spatial_scale, 8 | THCudaTensor * features, THCudaTensor * rois, THCudaTensor * output, THCudaIntTensor * argmax) 9 | { 10 | // Grab the input tensor 11 | float * data_flat = THCudaTensor_data(state, features); 12 | float * rois_flat = THCudaTensor_data(state, rois); 13 | 14 | float * output_flat = THCudaTensor_data(state, output); 15 | int * argmax_flat = THCudaIntTensor_data(state, argmax); 16 | 17 | // Number of ROIs 18 | int num_rois = THCudaTensor_size(state, rois, 0); 19 | int size_rois = THCudaTensor_size(state, rois, 1); 20 | if (size_rois != 5) 21 | { 22 | return 0; 23 | } 24 | 25 | // batch size 26 | // int batch_size = THCudaTensor_size(state, features, 0); 27 | // if (batch_size != 1) 28 | // { 29 | // return 0; 30 | // } 31 | // data height 32 | int data_height = THCudaTensor_size(state, features, 2); 33 | // data width 34 | int data_width = THCudaTensor_size(state, features, 3); 35 | // Number of channels 36 | int num_channels = THCudaTensor_size(state, features, 1); 37 | 38 | cudaStream_t stream = THCState_getCurrentStream(state); 39 | 40 | ROIPoolForwardLaucher( 41 | data_flat, spatial_scale, num_rois, data_height, 42 | data_width, num_channels, pooled_height, 43 | pooled_width, rois_flat, 44 | output_flat, argmax_flat, stream); 45 | 46 | return 1; 47 | } 48 | 49 | int roi_pooling_backward_cuda(int pooled_height, int pooled_width, float spatial_scale, 50 | THCudaTensor * top_grad, THCudaTensor * rois, THCudaTensor * bottom_grad, THCudaIntTensor * argmax) 51 | { 52 | // Grab the input tensor 53 | float * top_grad_flat = THCudaTensor_data(state, top_grad); 54 | float * rois_flat = THCudaTensor_data(state, rois); 55 | 56 | float * bottom_grad_flat = THCudaTensor_data(state, bottom_grad); 57 | int * argmax_flat = THCudaIntTensor_data(state, argmax); 58 | 59 | // Number of ROIs 60 | int num_rois = THCudaTensor_size(state, rois, 0); 61 | int size_rois = THCudaTensor_size(state, rois, 1); 62 | if (size_rois != 5) 63 | { 64 | return 0; 65 | } 66 | 67 | // batch size 68 | int batch_size = THCudaTensor_size(state, bottom_grad, 0); 69 | // if (batch_size != 1) 70 | // { 71 | // return 0; 72 | // } 73 | // data height 74 | int data_height = THCudaTensor_size(state, bottom_grad, 2); 75 | // data width 76 | int data_width = THCudaTensor_size(state, bottom_grad, 3); 77 | // Number of channels 78 | int num_channels = THCudaTensor_size(state, bottom_grad, 1); 79 | 80 | cudaStream_t stream = THCState_getCurrentStream(state); 81 | ROIPoolBackwardLaucher( 82 | top_grad_flat, spatial_scale, batch_size, num_rois, data_height, 83 | data_width, num_channels, pooled_height, 84 | pooled_width, rois_flat, 85 | bottom_grad_flat, argmax_flat, stream); 86 | 87 | return 1; 88 | } 89 | -------------------------------------------------------------------------------- /lib/model/roi_pooling/src/roi_pooling_cuda.h: -------------------------------------------------------------------------------- 1 | int roi_pooling_forward_cuda(int pooled_height, int pooled_width, float spatial_scale, 2 | THCudaTensor * features, THCudaTensor * rois, THCudaTensor * output, THCudaIntTensor * argmax); 3 | 4 | int roi_pooling_backward_cuda(int pooled_height, int pooled_width, float spatial_scale, 5 | THCudaTensor * top_grad, THCudaTensor * rois, THCudaTensor * bottom_grad, THCudaIntTensor * argmax); -------------------------------------------------------------------------------- /lib/model/roi_pooling/src/roi_pooling_kernel.cu: -------------------------------------------------------------------------------- 1 | // #ifdef __cplusplus 2 | // extern "C" { 3 | // #endif 4 | 5 | #include 6 | #include 7 | #include 8 | #include 9 | #include "roi_pooling_kernel.h" 10 | 11 | 12 | #define DIVUP(m, n) ((m) / (m) + ((m) % (n) > 0)) 13 | 14 | #define CUDA_1D_KERNEL_LOOP(i, n) \ 15 | for (int i = blockIdx.x * blockDim.x + threadIdx.x; i < n; \ 16 | i += blockDim.x * gridDim.x) 17 | 18 | // CUDA: grid stride looping 19 | #define CUDA_KERNEL_LOOP(i, n) \ 20 | for (int i = blockIdx.x * blockDim.x + threadIdx.x; \ 21 | i < (n); \ 22 | i += blockDim.x * gridDim.x) 23 | 24 | __global__ void ROIPoolForward(const int nthreads, const float* bottom_data, 25 | const float spatial_scale, const int height, const int width, 26 | const int channels, const int pooled_height, const int pooled_width, 27 | const float* bottom_rois, float* top_data, int* argmax_data) 28 | { 29 | CUDA_KERNEL_LOOP(index, nthreads) 30 | { 31 | // (n, c, ph, pw) is an element in the pooled output 32 | // int n = index; 33 | // int pw = n % pooled_width; 34 | // n /= pooled_width; 35 | // int ph = n % pooled_height; 36 | // n /= pooled_height; 37 | // int c = n % channels; 38 | // n /= channels; 39 | int pw = index % pooled_width; 40 | int ph = (index / pooled_width) % pooled_height; 41 | int c = (index / pooled_width / pooled_height) % channels; 42 | int n = index / pooled_width / pooled_height / channels; 43 | 44 | // bottom_rois += n * 5; 45 | int roi_batch_ind = bottom_rois[n * 5 + 0]; 46 | int roi_start_w = round(bottom_rois[n * 5 + 1] * spatial_scale); 47 | int roi_start_h = round(bottom_rois[n * 5 + 2] * spatial_scale); 48 | int roi_end_w = round(bottom_rois[n * 5 + 3] * spatial_scale); 49 | int roi_end_h = round(bottom_rois[n * 5 + 4] * spatial_scale); 50 | 51 | // Force malformed ROIs to be 1x1 52 | int roi_width = fmaxf(roi_end_w - roi_start_w + 1, 1); 53 | int roi_height = fmaxf(roi_end_h - roi_start_h + 1, 1); 54 | float bin_size_h = (float)(roi_height) / (float)(pooled_height); 55 | float bin_size_w = (float)(roi_width) / (float)(pooled_width); 56 | 57 | int hstart = (int)(floor((float)(ph) * bin_size_h)); 58 | int wstart = (int)(floor((float)(pw) * bin_size_w)); 59 | int hend = (int)(ceil((float)(ph + 1) * bin_size_h)); 60 | int wend = (int)(ceil((float)(pw + 1) * bin_size_w)); 61 | 62 | // Add roi offsets and clip to input boundaries 63 | hstart = fminf(fmaxf(hstart + roi_start_h, 0), height); 64 | hend = fminf(fmaxf(hend + roi_start_h, 0), height); 65 | wstart = fminf(fmaxf(wstart + roi_start_w, 0), width); 66 | wend = fminf(fmaxf(wend + roi_start_w, 0), width); 67 | bool is_empty = (hend <= hstart) || (wend <= wstart); 68 | 69 | // Define an empty pooling region to be zero 70 | float maxval = is_empty ? 0 : -FLT_MAX; 71 | // If nothing is pooled, argmax = -1 causes nothing to be backprop'd 72 | int maxidx = -1; 73 | // bottom_data += roi_batch_ind * channels * height * width; 74 | 75 | int bottom_data_batch_offset = roi_batch_ind * channels * height * width; 76 | int bottom_data_offset = bottom_data_batch_offset + c * height * width; 77 | 78 | for (int h = hstart; h < hend; ++h) { 79 | for (int w = wstart; w < wend; ++w) { 80 | // int bottom_index = (h * width + w) * channels + c; 81 | // int bottom_index = (c * height + h) * width + w; 82 | int bottom_index = h * width + w; 83 | if (bottom_data[bottom_data_offset + bottom_index] > maxval) { 84 | maxval = bottom_data[bottom_data_offset + bottom_index]; 85 | maxidx = bottom_data_offset + bottom_index; 86 | } 87 | } 88 | } 89 | top_data[index] = maxval; 90 | if (argmax_data != NULL) 91 | argmax_data[index] = maxidx; 92 | } 93 | } 94 | 95 | int ROIPoolForwardLaucher( 96 | const float* bottom_data, const float spatial_scale, const int num_rois, const int height, 97 | const int width, const int channels, const int pooled_height, 98 | const int pooled_width, const float* bottom_rois, 99 | float* top_data, int* argmax_data, cudaStream_t stream) 100 | { 101 | const int kThreadsPerBlock = 1024; 102 | int output_size = num_rois * pooled_height * pooled_width * channels; 103 | cudaError_t err; 104 | 105 | ROIPoolForward<<<(output_size + kThreadsPerBlock - 1) / kThreadsPerBlock, kThreadsPerBlock, 0, stream>>>( 106 | output_size, bottom_data, spatial_scale, height, width, channels, pooled_height, 107 | pooled_width, bottom_rois, top_data, argmax_data); 108 | 109 | // dim3 blocks(DIVUP(output_size, kThreadsPerBlock), 110 | // DIVUP(output_size, kThreadsPerBlock)); 111 | // dim3 threads(kThreadsPerBlock); 112 | // 113 | // ROIPoolForward<<>>( 114 | // output_size, bottom_data, spatial_scale, height, width, channels, pooled_height, 115 | // pooled_width, bottom_rois, top_data, argmax_data); 116 | 117 | err = cudaGetLastError(); 118 | if(cudaSuccess != err) 119 | { 120 | fprintf( stderr, "cudaCheckError() failed : %s\n", cudaGetErrorString( err ) ); 121 | exit( -1 ); 122 | } 123 | 124 | return 1; 125 | } 126 | 127 | 128 | __global__ void ROIPoolBackward(const int nthreads, const float* top_diff, 129 | const int* argmax_data, const int num_rois, const float spatial_scale, 130 | const int height, const int width, const int channels, 131 | const int pooled_height, const int pooled_width, float* bottom_diff, 132 | const float* bottom_rois) { 133 | CUDA_1D_KERNEL_LOOP(index, nthreads) 134 | { 135 | 136 | // (n, c, ph, pw) is an element in the pooled output 137 | int n = index; 138 | int w = n % width; 139 | n /= width; 140 | int h = n % height; 141 | n /= height; 142 | int c = n % channels; 143 | n /= channels; 144 | 145 | float gradient = 0; 146 | // Accumulate gradient over all ROIs that pooled this element 147 | for (int roi_n = 0; roi_n < num_rois; ++roi_n) 148 | { 149 | const float* offset_bottom_rois = bottom_rois + roi_n * 5; 150 | int roi_batch_ind = offset_bottom_rois[0]; 151 | // Skip if ROI's batch index doesn't match n 152 | if (n != roi_batch_ind) { 153 | continue; 154 | } 155 | 156 | int roi_start_w = round(offset_bottom_rois[1] * spatial_scale); 157 | int roi_start_h = round(offset_bottom_rois[2] * spatial_scale); 158 | int roi_end_w = round(offset_bottom_rois[3] * spatial_scale); 159 | int roi_end_h = round(offset_bottom_rois[4] * spatial_scale); 160 | 161 | // Skip if ROI doesn't include (h, w) 162 | const bool in_roi = (w >= roi_start_w && w <= roi_end_w && 163 | h >= roi_start_h && h <= roi_end_h); 164 | if (!in_roi) { 165 | continue; 166 | } 167 | 168 | int offset = roi_n * pooled_height * pooled_width * channels; 169 | const float* offset_top_diff = top_diff + offset; 170 | const int* offset_argmax_data = argmax_data + offset; 171 | 172 | // Compute feasible set of pooled units that could have pooled 173 | // this bottom unit 174 | 175 | // Force malformed ROIs to be 1x1 176 | int roi_width = fmaxf(roi_end_w - roi_start_w + 1, 1); 177 | int roi_height = fmaxf(roi_end_h - roi_start_h + 1, 1); 178 | 179 | float bin_size_h = (float)(roi_height) / (float)(pooled_height); 180 | float bin_size_w = (float)(roi_width) / (float)(pooled_width); 181 | 182 | int phstart = floor((float)(h - roi_start_h) / bin_size_h); 183 | int phend = ceil((float)(h - roi_start_h + 1) / bin_size_h); 184 | int pwstart = floor((float)(w - roi_start_w) / bin_size_w); 185 | int pwend = ceil((float)(w - roi_start_w + 1) / bin_size_w); 186 | 187 | phstart = fminf(fmaxf(phstart, 0), pooled_height); 188 | phend = fminf(fmaxf(phend, 0), pooled_height); 189 | pwstart = fminf(fmaxf(pwstart, 0), pooled_width); 190 | pwend = fminf(fmaxf(pwend, 0), pooled_width); 191 | 192 | for (int ph = phstart; ph < phend; ++ph) { 193 | for (int pw = pwstart; pw < pwend; ++pw) { 194 | if (offset_argmax_data[(c * pooled_height + ph) * pooled_width + pw] == index) 195 | { 196 | gradient += offset_top_diff[(c * pooled_height + ph) * pooled_width + pw]; 197 | } 198 | } 199 | } 200 | } 201 | bottom_diff[index] = gradient; 202 | } 203 | } 204 | 205 | int ROIPoolBackwardLaucher(const float* top_diff, const float spatial_scale, const int batch_size, const int num_rois, 206 | const int height, const int width, const int channels, const int pooled_height, 207 | const int pooled_width, const float* bottom_rois, 208 | float* bottom_diff, const int* argmax_data, cudaStream_t stream) 209 | { 210 | const int kThreadsPerBlock = 1024; 211 | int output_size = batch_size * height * width * channels; 212 | cudaError_t err; 213 | 214 | ROIPoolBackward<<<(output_size + kThreadsPerBlock - 1) / kThreadsPerBlock, kThreadsPerBlock, 0, stream>>>( 215 | output_size, top_diff, argmax_data, num_rois, spatial_scale, height, width, channels, pooled_height, 216 | pooled_width, bottom_diff, bottom_rois); 217 | 218 | // dim3 blocks(DIVUP(output_size, kThreadsPerBlock), 219 | // DIVUP(output_size, kThreadsPerBlock)); 220 | // dim3 threads(kThreadsPerBlock); 221 | // 222 | // ROIPoolBackward<<>>( 223 | // output_size, top_diff, argmax_data, num_rois, spatial_scale, height, width, channels, pooled_height, 224 | // pooled_width, bottom_diff, bottom_rois); 225 | 226 | err = cudaGetLastError(); 227 | if(cudaSuccess != err) 228 | { 229 | fprintf( stderr, "cudaCheckError() failed : %s\n", cudaGetErrorString( err ) ); 230 | exit( -1 ); 231 | } 232 | 233 | return 1; 234 | } 235 | 236 | 237 | // #ifdef __cplusplus 238 | // } 239 | // #endif 240 | -------------------------------------------------------------------------------- /lib/model/roi_pooling/src/roi_pooling_kernel.h: -------------------------------------------------------------------------------- 1 | #ifndef _ROI_POOLING_KERNEL 2 | #define _ROI_POOLING_KERNEL 3 | 4 | #ifdef __cplusplus 5 | extern "C" { 6 | #endif 7 | 8 | int ROIPoolForwardLaucher( 9 | const float* bottom_data, const float spatial_scale, const int num_rois, const int height, 10 | const int width, const int channels, const int pooled_height, 11 | const int pooled_width, const float* bottom_rois, 12 | float* top_data, int* argmax_data, cudaStream_t stream); 13 | 14 | 15 | int ROIPoolBackwardLaucher(const float* top_diff, const float spatial_scale, const int batch_size, const int num_rois, 16 | const int height, const int width, const int channels, const int pooled_height, 17 | const int pooled_width, const float* bottom_rois, 18 | float* bottom_diff, const int* argmax_data, cudaStream_t stream); 19 | 20 | #ifdef __cplusplus 21 | } 22 | #endif 23 | 24 | #endif 25 | 26 | -------------------------------------------------------------------------------- /lib/model/rpn/__init__.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/basiclab/DA-OD-MEAA-PyTorch/ab590ddacdd06886fdbd56f7c4f59cd36167d100/lib/model/rpn/__init__.py -------------------------------------------------------------------------------- /lib/model/rpn/anchor_target_layer.py: -------------------------------------------------------------------------------- 1 | from __future__ import absolute_import 2 | # -------------------------------------------------------- 3 | # Faster R-CNN 4 | # Copyright (c) 2015 Microsoft 5 | # Licensed under The MIT License [see LICENSE for details] 6 | # Written by Ross Girshick and Sean Bell 7 | # -------------------------------------------------------- 8 | # -------------------------------------------------------- 9 | # Reorganized and modified by Jianwei Yang and Jiasen Lu 10 | # -------------------------------------------------------- 11 | 12 | import torch 13 | import torch.nn as nn 14 | import numpy as np 15 | import numpy.random as npr 16 | 17 | from model.utils.config import cfg 18 | from .generate_anchors import generate_anchors 19 | from .bbox_transform import clip_boxes, bbox_overlaps_batch, bbox_transform_batch 20 | 21 | import pdb 22 | 23 | DEBUG = False 24 | 25 | try: 26 | long # Python 2 27 | except NameError: 28 | long = int # Python 3 29 | 30 | 31 | class _AnchorTargetLayer(nn.Module): 32 | """ 33 | Assign anchors to ground-truth targets. Produces anchor classification 34 | labels and bounding-box regression targets. 35 | """ 36 | def __init__(self, feat_stride, scales, ratios): 37 | super(_AnchorTargetLayer, self).__init__() 38 | 39 | self._feat_stride = feat_stride 40 | self._scales = scales 41 | anchor_scales = scales 42 | self._anchors = torch.from_numpy(generate_anchors(scales=np.array(anchor_scales), ratios=np.array(ratios))).float() 43 | self._num_anchors = self._anchors.size(0) 44 | 45 | # allow boxes to sit over the edge by a small amount 46 | self._allowed_border = 0 # default is 0 47 | 48 | def forward(self, input): 49 | # Algorithm: 50 | # 51 | # for each (H, W) location i 52 | # generate 9 anchor boxes centered on cell i 53 | # apply predicted bbox deltas at cell i to each of the 9 anchors 54 | # filter out-of-image anchors 55 | 56 | rpn_cls_score = input[0] 57 | gt_boxes = input[1] 58 | im_info = input[2] 59 | num_boxes = input[3] 60 | 61 | # map of shape (..., H, W) 62 | height, width = rpn_cls_score.size(2), rpn_cls_score.size(3) 63 | 64 | batch_size = gt_boxes.size(0) 65 | 66 | feat_height, feat_width = rpn_cls_score.size(2), rpn_cls_score.size(3) 67 | shift_x = np.arange(0, feat_width) * self._feat_stride 68 | shift_y = np.arange(0, feat_height) * self._feat_stride 69 | shift_x, shift_y = np.meshgrid(shift_x, shift_y) 70 | shifts = torch.from_numpy(np.vstack((shift_x.ravel(), shift_y.ravel(), 71 | shift_x.ravel(), shift_y.ravel())).transpose()) 72 | shifts = shifts.contiguous().type_as(rpn_cls_score).float() 73 | 74 | A = self._num_anchors 75 | K = shifts.size(0) 76 | 77 | self._anchors = self._anchors.type_as(gt_boxes) # move to specific gpu. 78 | all_anchors = self._anchors.view(1, A, 4) + shifts.view(K, 1, 4) 79 | all_anchors = all_anchors.view(K * A, 4) 80 | 81 | total_anchors = int(K * A) 82 | 83 | keep = ((all_anchors[:, 0] >= -self._allowed_border) & 84 | (all_anchors[:, 1] >= -self._allowed_border) & 85 | (all_anchors[:, 2] < long(im_info[0][1]) + self._allowed_border) & 86 | (all_anchors[:, 3] < long(im_info[0][0]) + self._allowed_border)) 87 | 88 | inds_inside = torch.nonzero(keep).view(-1) 89 | 90 | # keep only inside anchors 91 | anchors = all_anchors[inds_inside, :] 92 | 93 | # label: 1 is positive, 0 is negative, -1 is dont care 94 | labels = gt_boxes.new(batch_size, inds_inside.size(0)).fill_(-1) 95 | bbox_inside_weights = gt_boxes.new(batch_size, inds_inside.size(0)).zero_() 96 | bbox_outside_weights = gt_boxes.new(batch_size, inds_inside.size(0)).zero_() 97 | 98 | overlaps = bbox_overlaps_batch(anchors, gt_boxes) 99 | 100 | max_overlaps, argmax_overlaps = torch.max(overlaps, 2) 101 | gt_max_overlaps, _ = torch.max(overlaps, 1) 102 | 103 | if not cfg.TRAIN.RPN_CLOBBER_POSITIVES: 104 | labels[max_overlaps < cfg.TRAIN.RPN_NEGATIVE_OVERLAP] = 0 105 | 106 | gt_max_overlaps[gt_max_overlaps==0] = 1e-5 107 | keep = torch.sum(overlaps.eq(gt_max_overlaps.view(batch_size,1,-1).expand_as(overlaps)), 2) 108 | 109 | if torch.sum(keep) > 0: 110 | labels[keep>0] = 1 111 | 112 | # fg label: above threshold IOU 113 | labels[max_overlaps >= cfg.TRAIN.RPN_POSITIVE_OVERLAP] = 1 114 | 115 | if cfg.TRAIN.RPN_CLOBBER_POSITIVES: 116 | labels[max_overlaps < cfg.TRAIN.RPN_NEGATIVE_OVERLAP] = 0 117 | 118 | num_fg = int(cfg.TRAIN.RPN_FG_FRACTION * cfg.TRAIN.RPN_BATCHSIZE) 119 | 120 | sum_fg = torch.sum((labels == 1).int(), 1) 121 | sum_bg = torch.sum((labels == 0).int(), 1) 122 | 123 | for i in range(batch_size): 124 | # subsample positive labels if we have too many 125 | if sum_fg[i] > num_fg: 126 | fg_inds = torch.nonzero(labels[i] == 1).view(-1) 127 | # torch.randperm seems has a bug on multi-gpu setting that cause the segfault. 128 | # See https://github.com/pytorch/pytorch/issues/1868 for more details. 129 | # use numpy instead. 130 | #rand_num = torch.randperm(fg_inds.size(0)).type_as(gt_boxes).long() 131 | rand_num = torch.from_numpy(np.random.permutation(fg_inds.size(0))).type_as(gt_boxes).long() 132 | disable_inds = fg_inds[rand_num[:fg_inds.size(0)-num_fg]] 133 | labels[i][disable_inds] = -1 134 | 135 | # num_bg = cfg.TRAIN.RPN_BATCHSIZE - sum_fg[i] 136 | num_bg = cfg.TRAIN.RPN_BATCHSIZE - torch.sum((labels == 1).int(), 1)[i] 137 | 138 | # subsample negative labels if we have too many 139 | if sum_bg[i] > num_bg: 140 | bg_inds = torch.nonzero(labels[i] == 0).view(-1) 141 | #rand_num = torch.randperm(bg_inds.size(0)).type_as(gt_boxes).long() 142 | 143 | rand_num = torch.from_numpy(np.random.permutation(bg_inds.size(0))).type_as(gt_boxes).long() 144 | disable_inds = bg_inds[rand_num[:bg_inds.size(0)-num_bg]] 145 | labels[i][disable_inds] = -1 146 | 147 | offset = torch.arange(0, batch_size)*gt_boxes.size(1) 148 | 149 | argmax_overlaps = argmax_overlaps + offset.view(batch_size, 1).type_as(argmax_overlaps) 150 | bbox_targets = _compute_targets_batch(anchors, gt_boxes.view(-1,5)[argmax_overlaps.view(-1), :].view(batch_size, -1, 5)) 151 | 152 | # use a single value instead of 4 values for easy index. 153 | bbox_inside_weights[labels==1] = cfg.TRAIN.RPN_BBOX_INSIDE_WEIGHTS[0] 154 | 155 | if cfg.TRAIN.RPN_POSITIVE_WEIGHT < 0: 156 | num_examples = torch.sum(labels[i] >= 0) 157 | positive_weights = 1.0 / num_examples.item() 158 | negative_weights = 1.0 / num_examples.item() 159 | else: 160 | assert ((cfg.TRAIN.RPN_POSITIVE_WEIGHT > 0) & 161 | (cfg.TRAIN.RPN_POSITIVE_WEIGHT < 1)) 162 | 163 | bbox_outside_weights[labels == 1] = positive_weights 164 | bbox_outside_weights[labels == 0] = negative_weights 165 | 166 | labels = _unmap(labels, total_anchors, inds_inside, batch_size, fill=-1) 167 | bbox_targets = _unmap(bbox_targets, total_anchors, inds_inside, batch_size, fill=0) 168 | bbox_inside_weights = _unmap(bbox_inside_weights, total_anchors, inds_inside, batch_size, fill=0) 169 | bbox_outside_weights = _unmap(bbox_outside_weights, total_anchors, inds_inside, batch_size, fill=0) 170 | 171 | outputs = [] 172 | 173 | labels = labels.view(batch_size, height, width, A).permute(0,3,1,2).contiguous() 174 | labels = labels.view(batch_size, 1, A * height, width) 175 | outputs.append(labels) 176 | 177 | bbox_targets = bbox_targets.view(batch_size, height, width, A*4).permute(0,3,1,2).contiguous() 178 | outputs.append(bbox_targets) 179 | 180 | anchors_count = bbox_inside_weights.size(1) 181 | bbox_inside_weights = bbox_inside_weights.view(batch_size,anchors_count,1).expand(batch_size, anchors_count, 4) 182 | 183 | bbox_inside_weights = bbox_inside_weights.contiguous().view(batch_size, height, width, 4*A)\ 184 | .permute(0,3,1,2).contiguous() 185 | 186 | outputs.append(bbox_inside_weights) 187 | 188 | bbox_outside_weights = bbox_outside_weights.view(batch_size,anchors_count,1).expand(batch_size, anchors_count, 4) 189 | bbox_outside_weights = bbox_outside_weights.contiguous().view(batch_size, height, width, 4*A)\ 190 | .permute(0,3,1,2).contiguous() 191 | outputs.append(bbox_outside_weights) 192 | 193 | return outputs 194 | 195 | def backward(self, top, propagate_down, bottom): 196 | """This layer does not propagate gradients.""" 197 | pass 198 | 199 | def reshape(self, bottom, top): 200 | """Reshaping happens during the call to forward.""" 201 | pass 202 | 203 | def _unmap(data, count, inds, batch_size, fill=0): 204 | """ Unmap a subset of item (data) back to the original set of items (of 205 | size count) """ 206 | 207 | if data.dim() == 2: 208 | ret = torch.Tensor(batch_size, count).fill_(fill).type_as(data) 209 | ret[:, inds] = data 210 | else: 211 | ret = torch.Tensor(batch_size, count, data.size(2)).fill_(fill).type_as(data) 212 | ret[:, inds,:] = data 213 | return ret 214 | 215 | 216 | def _compute_targets_batch(ex_rois, gt_rois): 217 | """Compute bounding-box regression targets for an image.""" 218 | 219 | return bbox_transform_batch(ex_rois, gt_rois[:, :, :4]) 220 | -------------------------------------------------------------------------------- /lib/model/rpn/bbox_transform.py: -------------------------------------------------------------------------------- 1 | # -------------------------------------------------------- 2 | # Fast R-CNN 3 | # Copyright (c) 2015 Microsoft 4 | # Licensed under The MIT License [see LICENSE for details] 5 | # Written by Ross Girshick 6 | # -------------------------------------------------------- 7 | # -------------------------------------------------------- 8 | # Reorganized and modified by Jianwei Yang and Jiasen Lu 9 | # -------------------------------------------------------- 10 | 11 | import torch 12 | import numpy as np 13 | import pdb 14 | 15 | def bbox_transform(ex_rois, gt_rois): 16 | ex_widths = ex_rois[:, 2] - ex_rois[:, 0] + 1.0 17 | ex_heights = ex_rois[:, 3] - ex_rois[:, 1] + 1.0 18 | ex_ctr_x = ex_rois[:, 0] + 0.5 * ex_widths 19 | ex_ctr_y = ex_rois[:, 1] + 0.5 * ex_heights 20 | 21 | gt_widths = gt_rois[:, 2] - gt_rois[:, 0] + 1.0 22 | gt_heights = gt_rois[:, 3] - gt_rois[:, 1] + 1.0 23 | gt_ctr_x = gt_rois[:, 0] + 0.5 * gt_widths 24 | gt_ctr_y = gt_rois[:, 1] + 0.5 * gt_heights 25 | 26 | targets_dx = (gt_ctr_x - ex_ctr_x) / ex_widths 27 | targets_dy = (gt_ctr_y - ex_ctr_y) / ex_heights 28 | targets_dw = torch.log(gt_widths / ex_widths) 29 | targets_dh = torch.log(gt_heights / ex_heights) 30 | 31 | targets = torch.stack( 32 | (targets_dx, targets_dy, targets_dw, targets_dh),1) 33 | 34 | return targets 35 | 36 | def bbox_transform_batch(ex_rois, gt_rois): 37 | 38 | if ex_rois.dim() == 2: 39 | ex_widths = ex_rois[:, 2] - ex_rois[:, 0] + 1.0 40 | ex_heights = ex_rois[:, 3] - ex_rois[:, 1] + 1.0 41 | ex_ctr_x = ex_rois[:, 0] + 0.5 * ex_widths 42 | ex_ctr_y = ex_rois[:, 1] + 0.5 * ex_heights 43 | 44 | gt_widths = gt_rois[:, :, 2] - gt_rois[:, :, 0] + 1.0 45 | gt_heights = gt_rois[:, :, 3] - gt_rois[:, :, 1] + 1.0 46 | gt_ctr_x = gt_rois[:, :, 0] + 0.5 * gt_widths 47 | gt_ctr_y = gt_rois[:, :, 1] + 0.5 * gt_heights 48 | 49 | targets_dx = (gt_ctr_x - ex_ctr_x.view(1,-1).expand_as(gt_ctr_x)) / ex_widths 50 | targets_dy = (gt_ctr_y - ex_ctr_y.view(1,-1).expand_as(gt_ctr_y)) / ex_heights 51 | targets_dw = torch.log(gt_widths / ex_widths.view(1,-1).expand_as(gt_widths)) 52 | targets_dh = torch.log(gt_heights / ex_heights.view(1,-1).expand_as(gt_heights)) 53 | 54 | elif ex_rois.dim() == 3: 55 | ex_widths = ex_rois[:, :, 2] - ex_rois[:, :, 0] + 1.0 56 | ex_heights = ex_rois[:,:, 3] - ex_rois[:,:, 1] + 1.0 57 | ex_ctr_x = ex_rois[:, :, 0] + 0.5 * ex_widths 58 | ex_ctr_y = ex_rois[:, :, 1] + 0.5 * ex_heights 59 | 60 | gt_widths = gt_rois[:, :, 2] - gt_rois[:, :, 0] + 1.0 61 | gt_heights = gt_rois[:, :, 3] - gt_rois[:, :, 1] + 1.0 62 | gt_ctr_x = gt_rois[:, :, 0] + 0.5 * gt_widths 63 | gt_ctr_y = gt_rois[:, :, 1] + 0.5 * gt_heights 64 | 65 | targets_dx = (gt_ctr_x - ex_ctr_x) / ex_widths 66 | targets_dy = (gt_ctr_y - ex_ctr_y) / ex_heights 67 | targets_dw = torch.log(gt_widths / ex_widths) 68 | targets_dh = torch.log(gt_heights / ex_heights) 69 | else: 70 | raise ValueError('ex_roi input dimension is not correct.') 71 | 72 | targets = torch.stack( 73 | (targets_dx, targets_dy, targets_dw, targets_dh),2) 74 | 75 | return targets 76 | 77 | def bbox_transform_inv(boxes, deltas, batch_size): 78 | widths = boxes[:, :, 2] - boxes[:, :, 0] + 1.0 79 | heights = boxes[:, :, 3] - boxes[:, :, 1] + 1.0 80 | ctr_x = boxes[:, :, 0] + 0.5 * widths 81 | ctr_y = boxes[:, :, 1] + 0.5 * heights 82 | 83 | dx = deltas[:, :, 0::4] 84 | dy = deltas[:, :, 1::4] 85 | dw = deltas[:, :, 2::4] 86 | dh = deltas[:, :, 3::4] 87 | 88 | pred_ctr_x = dx * widths.unsqueeze(2) + ctr_x.unsqueeze(2) 89 | pred_ctr_y = dy * heights.unsqueeze(2) + ctr_y.unsqueeze(2) 90 | pred_w = torch.exp(dw) * widths.unsqueeze(2) 91 | pred_h = torch.exp(dh) * heights.unsqueeze(2) 92 | 93 | pred_boxes = deltas.clone() 94 | # x1 95 | pred_boxes[:, :, 0::4] = pred_ctr_x - 0.5 * pred_w 96 | # y1 97 | pred_boxes[:, :, 1::4] = pred_ctr_y - 0.5 * pred_h 98 | # x2 99 | pred_boxes[:, :, 2::4] = pred_ctr_x + 0.5 * pred_w 100 | # y2 101 | pred_boxes[:, :, 3::4] = pred_ctr_y + 0.5 * pred_h 102 | 103 | return pred_boxes 104 | 105 | def clip_boxes_batch(boxes, im_shape, batch_size): 106 | """ 107 | Clip boxes to image boundaries. 108 | """ 109 | num_rois = boxes.size(1) 110 | 111 | boxes[boxes < 0] = 0 112 | # batch_x = (im_shape[:,0]-1).view(batch_size, 1).expand(batch_size, num_rois) 113 | # batch_y = (im_shape[:,1]-1).view(batch_size, 1).expand(batch_size, num_rois) 114 | 115 | batch_x = im_shape[:, 1] - 1 116 | batch_y = im_shape[:, 0] - 1 117 | 118 | boxes[:,:,0][boxes[:,:,0] > batch_x] = batch_x 119 | boxes[:,:,1][boxes[:,:,1] > batch_y] = batch_y 120 | boxes[:,:,2][boxes[:,:,2] > batch_x] = batch_x 121 | boxes[:,:,3][boxes[:,:,3] > batch_y] = batch_y 122 | 123 | return boxes 124 | 125 | def clip_boxes(boxes, im_shape, batch_size): 126 | 127 | for i in range(batch_size): 128 | boxes[i,:,0::4].clamp_(0, im_shape[i, 1]-1) 129 | boxes[i,:,1::4].clamp_(0, im_shape[i, 0]-1) 130 | boxes[i,:,2::4].clamp_(0, im_shape[i, 1]-1) 131 | boxes[i,:,3::4].clamp_(0, im_shape[i, 0]-1) 132 | 133 | return boxes 134 | 135 | 136 | def bbox_overlaps(anchors, gt_boxes): 137 | """ 138 | anchors: (N, 4) ndarray of float 139 | gt_boxes: (K, 4) ndarray of float 140 | 141 | overlaps: (N, K) ndarray of overlap between boxes and query_boxes 142 | """ 143 | N = anchors.size(0) 144 | K = gt_boxes.size(0) 145 | 146 | gt_boxes_area = ((gt_boxes[:,2] - gt_boxes[:,0] + 1) * 147 | (gt_boxes[:,3] - gt_boxes[:,1] + 1)).view(1, K) 148 | 149 | anchors_area = ((anchors[:,2] - anchors[:,0] + 1) * 150 | (anchors[:,3] - anchors[:,1] + 1)).view(N, 1) 151 | 152 | boxes = anchors.view(N, 1, 4).expand(N, K, 4) 153 | query_boxes = gt_boxes.view(1, K, 4).expand(N, K, 4) 154 | 155 | iw = (torch.min(boxes[:,:,2], query_boxes[:,:,2]) - 156 | torch.max(boxes[:,:,0], query_boxes[:,:,0]) + 1) 157 | iw[iw < 0] = 0 158 | 159 | ih = (torch.min(boxes[:,:,3], query_boxes[:,:,3]) - 160 | torch.max(boxes[:,:,1], query_boxes[:,:,1]) + 1) 161 | ih[ih < 0] = 0 162 | 163 | ua = anchors_area + gt_boxes_area - (iw * ih) 164 | overlaps = iw * ih / ua 165 | 166 | return overlaps 167 | 168 | def bbox_overlaps_batch(anchors, gt_boxes): 169 | """ 170 | anchors: (N, 4) ndarray of float 171 | gt_boxes: (b, K, 5) ndarray of float 172 | 173 | overlaps: (N, K) ndarray of overlap between boxes and query_boxes 174 | """ 175 | batch_size = gt_boxes.size(0) 176 | 177 | 178 | if anchors.dim() == 2: 179 | 180 | N = anchors.size(0) 181 | K = gt_boxes.size(1) 182 | 183 | anchors = anchors.view(1, N, 4).expand(batch_size, N, 4).contiguous() 184 | gt_boxes = gt_boxes[:,:,:4].contiguous() 185 | 186 | 187 | gt_boxes_x = (gt_boxes[:,:,2] - gt_boxes[:,:,0] + 1) 188 | gt_boxes_y = (gt_boxes[:,:,3] - gt_boxes[:,:,1] + 1) 189 | gt_boxes_area = (gt_boxes_x * gt_boxes_y).view(batch_size, 1, K) 190 | 191 | anchors_boxes_x = (anchors[:,:,2] - anchors[:,:,0] + 1) 192 | anchors_boxes_y = (anchors[:,:,3] - anchors[:,:,1] + 1) 193 | anchors_area = (anchors_boxes_x * anchors_boxes_y).view(batch_size, N, 1) 194 | 195 | gt_area_zero = (gt_boxes_x == 1) & (gt_boxes_y == 1) 196 | anchors_area_zero = (anchors_boxes_x == 1) & (anchors_boxes_y == 1) 197 | 198 | boxes = anchors.view(batch_size, N, 1, 4).expand(batch_size, N, K, 4) 199 | query_boxes = gt_boxes.view(batch_size, 1, K, 4).expand(batch_size, N, K, 4) 200 | 201 | iw = (torch.min(boxes[:,:,:,2], query_boxes[:,:,:,2]) - 202 | torch.max(boxes[:,:,:,0], query_boxes[:,:,:,0]) + 1) 203 | iw[iw < 0] = 0 204 | 205 | ih = (torch.min(boxes[:,:,:,3], query_boxes[:,:,:,3]) - 206 | torch.max(boxes[:,:,:,1], query_boxes[:,:,:,1]) + 1) 207 | ih[ih < 0] = 0 208 | ua = anchors_area + gt_boxes_area - (iw * ih) 209 | overlaps = iw * ih / ua 210 | 211 | # mask the overlap here. 212 | overlaps.masked_fill_(gt_area_zero.view(batch_size, 1, K).expand(batch_size, N, K), 0) 213 | overlaps.masked_fill_(anchors_area_zero.view(batch_size, N, 1).expand(batch_size, N, K), -1) 214 | 215 | elif anchors.dim() == 3: 216 | N = anchors.size(1) 217 | K = gt_boxes.size(1) 218 | 219 | if anchors.size(2) == 4: 220 | anchors = anchors[:,:,:4].contiguous() 221 | else: 222 | anchors = anchors[:,:,1:5].contiguous() 223 | 224 | gt_boxes = gt_boxes[:,:,:4].contiguous() 225 | 226 | gt_boxes_x = (gt_boxes[:,:,2] - gt_boxes[:,:,0] + 1) 227 | gt_boxes_y = (gt_boxes[:,:,3] - gt_boxes[:,:,1] + 1) 228 | gt_boxes_area = (gt_boxes_x * gt_boxes_y).view(batch_size, 1, K) 229 | 230 | anchors_boxes_x = (anchors[:,:,2] - anchors[:,:,0] + 1) 231 | anchors_boxes_y = (anchors[:,:,3] - anchors[:,:,1] + 1) 232 | anchors_area = (anchors_boxes_x * anchors_boxes_y).view(batch_size, N, 1) 233 | 234 | gt_area_zero = (gt_boxes_x == 1) & (gt_boxes_y == 1) 235 | anchors_area_zero = (anchors_boxes_x == 1) & (anchors_boxes_y == 1) 236 | 237 | boxes = anchors.view(batch_size, N, 1, 4).expand(batch_size, N, K, 4) 238 | query_boxes = gt_boxes.view(batch_size, 1, K, 4).expand(batch_size, N, K, 4) 239 | 240 | iw = (torch.min(boxes[:,:,:,2], query_boxes[:,:,:,2]) - 241 | torch.max(boxes[:,:,:,0], query_boxes[:,:,:,0]) + 1) 242 | iw[iw < 0] = 0 243 | 244 | ih = (torch.min(boxes[:,:,:,3], query_boxes[:,:,:,3]) - 245 | torch.max(boxes[:,:,:,1], query_boxes[:,:,:,1]) + 1) 246 | ih[ih < 0] = 0 247 | ua = anchors_area + gt_boxes_area - (iw * ih) 248 | 249 | overlaps = iw * ih / ua 250 | 251 | # mask the overlap here. 252 | overlaps.masked_fill_(gt_area_zero.view(batch_size, 1, K).expand(batch_size, N, K), 0) 253 | overlaps.masked_fill_(anchors_area_zero.view(batch_size, N, 1).expand(batch_size, N, K), -1) 254 | else: 255 | raise ValueError('anchors input dimension is not correct.') 256 | 257 | return overlaps 258 | -------------------------------------------------------------------------------- /lib/model/rpn/generate_anchors.py: -------------------------------------------------------------------------------- 1 | from __future__ import print_function 2 | # -------------------------------------------------------- 3 | # Faster R-CNN 4 | # Copyright (c) 2015 Microsoft 5 | # Licensed under The MIT License [see LICENSE for details] 6 | # Written by Ross Girshick and Sean Bell 7 | # -------------------------------------------------------- 8 | 9 | import numpy as np 10 | import pdb 11 | 12 | # Verify that we compute the same anchors as Shaoqing's matlab implementation: 13 | # 14 | # >> load output/rpn_cachedir/faster_rcnn_VOC2007_ZF_stage1_rpn/anchors.mat 15 | # >> anchors 16 | # 17 | # anchors = 18 | # 19 | # -83 -39 100 56 20 | # -175 -87 192 104 21 | # -359 -183 376 200 22 | # -55 -55 72 72 23 | # -119 -119 136 136 24 | # -247 -247 264 264 25 | # -35 -79 52 96 26 | # -79 -167 96 184 27 | # -167 -343 184 360 28 | 29 | #array([[ -83., -39., 100., 56.], 30 | # [-175., -87., 192., 104.], 31 | # [-359., -183., 376., 200.], 32 | # [ -55., -55., 72., 72.], 33 | # [-119., -119., 136., 136.], 34 | # [-247., -247., 264., 264.], 35 | # [ -35., -79., 52., 96.], 36 | # [ -79., -167., 96., 184.], 37 | # [-167., -343., 184., 360.]]) 38 | 39 | try: 40 | xrange # Python 2 41 | except NameError: 42 | xrange = range # Python 3 43 | 44 | 45 | def generate_anchors(base_size=16, ratios=[0.5, 1, 2], 46 | scales=2**np.arange(3, 6)): 47 | """ 48 | Generate anchor (reference) windows by enumerating aspect ratios X 49 | scales wrt a reference (0, 0, 15, 15) window. 50 | """ 51 | 52 | base_anchor = np.array([1, 1, base_size, base_size]) - 1 53 | ratio_anchors = _ratio_enum(base_anchor, ratios) 54 | anchors = np.vstack([_scale_enum(ratio_anchors[i, :], scales) 55 | for i in xrange(ratio_anchors.shape[0])]) 56 | return anchors 57 | 58 | def _whctrs(anchor): 59 | """ 60 | Return width, height, x center, and y center for an anchor (window). 61 | """ 62 | 63 | w = anchor[2] - anchor[0] + 1 64 | h = anchor[3] - anchor[1] + 1 65 | x_ctr = anchor[0] + 0.5 * (w - 1) 66 | y_ctr = anchor[1] + 0.5 * (h - 1) 67 | return w, h, x_ctr, y_ctr 68 | 69 | def _mkanchors(ws, hs, x_ctr, y_ctr): 70 | """ 71 | Given a vector of widths (ws) and heights (hs) around a center 72 | (x_ctr, y_ctr), output a set of anchors (windows). 73 | """ 74 | 75 | ws = ws[:, np.newaxis] 76 | hs = hs[:, np.newaxis] 77 | anchors = np.hstack((x_ctr - 0.5 * (ws - 1), 78 | y_ctr - 0.5 * (hs - 1), 79 | x_ctr + 0.5 * (ws - 1), 80 | y_ctr + 0.5 * (hs - 1))) 81 | return anchors 82 | 83 | def _ratio_enum(anchor, ratios): 84 | """ 85 | Enumerate a set of anchors for each aspect ratio wrt an anchor. 86 | """ 87 | 88 | w, h, x_ctr, y_ctr = _whctrs(anchor) 89 | size = w * h 90 | size_ratios = size / ratios 91 | ws = np.round(np.sqrt(size_ratios)) 92 | hs = np.round(ws * ratios) 93 | anchors = _mkanchors(ws, hs, x_ctr, y_ctr) 94 | return anchors 95 | 96 | def _scale_enum(anchor, scales): 97 | """ 98 | Enumerate a set of anchors for each scale wrt an anchor. 99 | """ 100 | 101 | w, h, x_ctr, y_ctr = _whctrs(anchor) 102 | ws = w * scales 103 | hs = h * scales 104 | anchors = _mkanchors(ws, hs, x_ctr, y_ctr) 105 | return anchors 106 | 107 | if __name__ == '__main__': 108 | import time 109 | t = time.time() 110 | a = generate_anchors() 111 | print(time.time() - t) 112 | print(a) 113 | from IPython import embed; embed() 114 | -------------------------------------------------------------------------------- /lib/model/rpn/proposal_layer.py: -------------------------------------------------------------------------------- 1 | from __future__ import absolute_import 2 | # -------------------------------------------------------- 3 | # Faster R-CNN 4 | # Copyright (c) 2015 Microsoft 5 | # Licensed under The MIT License [see LICENSE for details] 6 | # Written by Ross Girshick and Sean Bell 7 | # -------------------------------------------------------- 8 | # -------------------------------------------------------- 9 | # Reorganized and modified by Jianwei Yang and Jiasen Lu 10 | # -------------------------------------------------------- 11 | 12 | import torch 13 | import torch.nn as nn 14 | import numpy as np 15 | import math 16 | import yaml 17 | from model.utils.config import cfg 18 | from .generate_anchors import generate_anchors 19 | from .bbox_transform import bbox_transform_inv, clip_boxes, clip_boxes_batch 20 | from model.nms.nms_wrapper import nms 21 | 22 | import pdb 23 | 24 | DEBUG = False 25 | 26 | class _ProposalLayer(nn.Module): 27 | """ 28 | Outputs object detection proposals by applying estimated bounding-box 29 | transformations to a set of regular boxes (called "anchors"). 30 | """ 31 | 32 | def __init__(self, feat_stride, scales, ratios): 33 | super(_ProposalLayer, self).__init__() 34 | 35 | self._feat_stride = feat_stride 36 | self._anchors = torch.from_numpy(generate_anchors(scales=np.array(scales), 37 | ratios=np.array(ratios))).float() 38 | self._num_anchors = self._anchors.size(0) 39 | 40 | # rois blob: holds R regions of interest, each is a 5-tuple 41 | # (n, x1, y1, x2, y2) specifying an image batch index n and a 42 | # rectangle (x1, y1, x2, y2) 43 | # top[0].reshape(1, 5) 44 | # 45 | # # scores blob: holds scores for R regions of interest 46 | # if len(top) > 1: 47 | # top[1].reshape(1, 1, 1, 1) 48 | 49 | def forward(self, input,target=False): 50 | 51 | # Algorithm: 52 | # 53 | # for each (H, W) location i 54 | # generate A anchor boxes centered on cell i 55 | # apply predicted bbox deltas at cell i to each of the A anchors 56 | # clip predicted boxes to image 57 | # remove predicted boxes with either height or width < threshold 58 | # sort all (proposal, score) pairs by score from highest to lowest 59 | # take top pre_nms_topN proposals before NMS 60 | # apply NMS with threshold 0.7 to remaining proposals 61 | # take after_nms_topN proposals after NMS 62 | # return the top proposals (-> RoIs top, scores top) 63 | 64 | 65 | # the first set of _num_anchors channels are bg probs 66 | # the second set are the fg probs 67 | scores = input[0][:, self._num_anchors:, :, :] 68 | bbox_deltas = input[1] 69 | im_info = input[2] 70 | cfg_key = input[3] 71 | 72 | pre_nms_topN = cfg[cfg_key].RPN_PRE_NMS_TOP_N 73 | post_nms_topN = cfg[cfg_key].RPN_POST_NMS_TOP_N 74 | if target: 75 | post_nms_topN = cfg[cfg_key].RPN_POST_NMS_TOP_N_TARGET 76 | nms_thresh = cfg[cfg_key].RPN_NMS_THRESH 77 | min_size = cfg[cfg_key].RPN_MIN_SIZE 78 | 79 | batch_size = bbox_deltas.size(0) 80 | 81 | feat_height, feat_width = scores.size(2), scores.size(3) 82 | shift_x = np.arange(0, feat_width) * self._feat_stride 83 | shift_y = np.arange(0, feat_height) * self._feat_stride 84 | shift_x, shift_y = np.meshgrid(shift_x, shift_y) 85 | shifts = torch.from_numpy(np.vstack((shift_x.ravel(), shift_y.ravel(), 86 | shift_x.ravel(), shift_y.ravel())).transpose()) 87 | shifts = shifts.contiguous().type_as(scores).float() 88 | 89 | A = self._num_anchors 90 | K = shifts.size(0) 91 | 92 | self._anchors = self._anchors.type_as(scores) 93 | # anchors = self._anchors.view(1, A, 4) + shifts.view(1, K, 4).permute(1, 0, 2).contiguous() 94 | anchors = self._anchors.view(1, A, 4) + shifts.view(K, 1, 4) 95 | anchors = anchors.view(1, K * A, 4).expand(batch_size, K * A, 4) 96 | 97 | # Transpose and reshape predicted bbox transformations to get them 98 | # into the same order as the anchors: 99 | 100 | bbox_deltas = bbox_deltas.permute(0, 2, 3, 1).contiguous() 101 | bbox_deltas = bbox_deltas.view(batch_size, -1, 4) 102 | 103 | # Same story for the scores: 104 | scores = scores.permute(0, 2, 3, 1).contiguous() 105 | scores = scores.view(batch_size, -1) 106 | 107 | # Convert anchors into proposals via bbox transformations 108 | proposals = bbox_transform_inv(anchors, bbox_deltas, batch_size) 109 | 110 | # 2. clip predicted boxes to image 111 | proposals = clip_boxes(proposals, im_info, batch_size) 112 | # proposals = clip_boxes_batch(proposals, im_info, batch_size) 113 | 114 | # assign the score to 0 if it's non keep. 115 | # keep = self._filter_boxes(proposals, min_size * im_info[:, 2]) 116 | 117 | # trim keep index to make it euqal over batch 118 | # keep_idx = torch.cat(tuple(keep_idx), 0) 119 | 120 | # scores_keep = scores.view(-1)[keep_idx].view(batch_size, trim_size) 121 | # proposals_keep = proposals.view(-1, 4)[keep_idx, :].contiguous().view(batch_size, trim_size, 4) 122 | 123 | # _, order = torch.sort(scores_keep, 1, True) 124 | 125 | scores_keep = scores 126 | proposals_keep = proposals 127 | _, order = torch.sort(scores_keep, 1, True) 128 | 129 | output = scores.new(batch_size, post_nms_topN, 5).zero_() 130 | for i in range(batch_size): 131 | # # 3. remove predicted boxes with either height or width < threshold 132 | # # (NOTE: convert min_size to input image scale stored in im_info[2]) 133 | proposals_single = proposals_keep[i] 134 | scores_single = scores_keep[i] 135 | 136 | # # 4. sort all (proposal, score) pairs by score from highest to lowest 137 | # # 5. take top pre_nms_topN (e.g. 6000) 138 | order_single = order[i] 139 | 140 | if pre_nms_topN > 0 and pre_nms_topN < scores_keep.numel(): 141 | order_single = order_single[:pre_nms_topN] 142 | 143 | proposals_single = proposals_single[order_single, :] 144 | scores_single = scores_single[order_single].view(-1,1) 145 | 146 | # 6. apply nms (e.g. threshold = 0.7) 147 | # 7. take after_nms_topN (e.g. 300) 148 | # 8. return the top proposals (-> RoIs top) 149 | 150 | keep_idx_i = nms(torch.cat((proposals_single, scores_single), 1), nms_thresh, force_cpu=not cfg.USE_GPU_NMS) 151 | keep_idx_i = keep_idx_i.long().view(-1) 152 | 153 | if post_nms_topN > 0: 154 | keep_idx_i = keep_idx_i[:post_nms_topN] 155 | proposals_single = proposals_single[keep_idx_i, :] 156 | scores_single = scores_single[keep_idx_i, :] 157 | 158 | # padding 0 at the end. 159 | num_proposal = proposals_single.size(0) 160 | output[i,:,0] = i 161 | output[i,:num_proposal,1:] = proposals_single 162 | 163 | return output 164 | 165 | def backward(self, top, propagate_down, bottom): 166 | """This layer does not propagate gradients.""" 167 | pass 168 | 169 | def reshape(self, bottom, top): 170 | """Reshaping happens during the call to forward.""" 171 | pass 172 | 173 | def _filter_boxes(self, boxes, min_size): 174 | """Remove all boxes with any side smaller than min_size.""" 175 | ws = boxes[:, :, 2] - boxes[:, :, 0] + 1 176 | hs = boxes[:, :, 3] - boxes[:, :, 1] + 1 177 | keep = ((ws >= min_size.view(-1,1).expand_as(ws)) & (hs >= min_size.view(-1,1).expand_as(hs))) 178 | return keep 179 | -------------------------------------------------------------------------------- /lib/model/rpn/proposal_target_layer_cascade.py: -------------------------------------------------------------------------------- 1 | from __future__ import absolute_import 2 | # -------------------------------------------------------- 3 | # Faster R-CNN 4 | # Copyright (c) 2015 Microsoft 5 | # Licensed under The MIT License [see LICENSE for details] 6 | # Written by Ross Girshick and Sean Bell 7 | # -------------------------------------------------------- 8 | # -------------------------------------------------------- 9 | # Reorganized and modified by Jianwei Yang and Jiasen Lu 10 | # -------------------------------------------------------- 11 | 12 | import torch 13 | import torch.nn as nn 14 | import numpy as np 15 | import numpy.random as npr 16 | from ..utils.config import cfg 17 | from .bbox_transform import bbox_overlaps_batch, bbox_transform_batch 18 | import pdb 19 | 20 | class _ProposalTargetLayer(nn.Module): 21 | """ 22 | Assign object detection proposals to ground-truth targets. Produces proposal 23 | classification labels and bounding-box regression targets. 24 | """ 25 | 26 | def __init__(self, nclasses): 27 | super(_ProposalTargetLayer, self).__init__() 28 | self._num_classes = nclasses 29 | self.BBOX_NORMALIZE_MEANS = torch.FloatTensor(cfg.TRAIN.BBOX_NORMALIZE_MEANS) 30 | self.BBOX_NORMALIZE_STDS = torch.FloatTensor(cfg.TRAIN.BBOX_NORMALIZE_STDS) 31 | self.BBOX_INSIDE_WEIGHTS = torch.FloatTensor(cfg.TRAIN.BBOX_INSIDE_WEIGHTS) 32 | 33 | def forward(self, all_rois, gt_boxes, num_boxes): 34 | 35 | self.BBOX_NORMALIZE_MEANS = self.BBOX_NORMALIZE_MEANS.type_as(gt_boxes) 36 | self.BBOX_NORMALIZE_STDS = self.BBOX_NORMALIZE_STDS.type_as(gt_boxes) 37 | self.BBOX_INSIDE_WEIGHTS = self.BBOX_INSIDE_WEIGHTS.type_as(gt_boxes) 38 | 39 | gt_boxes_append = gt_boxes.new(gt_boxes.size()).zero_() 40 | gt_boxes_append[:,:,1:5] = gt_boxes[:,:,:4] 41 | 42 | # Include ground-truth boxes in the set of candidate rois 43 | all_rois = torch.cat([all_rois, gt_boxes_append], 1) 44 | 45 | num_images = 1 46 | rois_per_image = int(cfg.TRAIN.BATCH_SIZE / num_images) 47 | fg_rois_per_image = int(np.round(cfg.TRAIN.FG_FRACTION * rois_per_image)) 48 | fg_rois_per_image = 1 if fg_rois_per_image == 0 else fg_rois_per_image 49 | 50 | labels, rois, bbox_targets, bbox_inside_weights = self._sample_rois_pytorch( 51 | all_rois, gt_boxes, fg_rois_per_image, 52 | rois_per_image, self._num_classes) 53 | 54 | bbox_outside_weights = (bbox_inside_weights > 0).float() 55 | 56 | return rois, labels, bbox_targets, bbox_inside_weights, bbox_outside_weights 57 | 58 | def backward(self, top, propagate_down, bottom): 59 | """This layer does not propagate gradients.""" 60 | pass 61 | 62 | def reshape(self, bottom, top): 63 | """Reshaping happens during the call to forward.""" 64 | pass 65 | 66 | def _get_bbox_regression_labels_pytorch(self, bbox_target_data, labels_batch, num_classes): 67 | """Bounding-box regression targets (bbox_target_data) are stored in a 68 | compact form b x N x (class, tx, ty, tw, th) 69 | 70 | This function expands those targets into the 4-of-4*K representation used 71 | by the network (i.e. only one class has non-zero targets). 72 | 73 | Returns: 74 | bbox_target (ndarray): b x N x 4K blob of regression targets 75 | bbox_inside_weights (ndarray): b x N x 4K blob of loss weights 76 | """ 77 | batch_size = labels_batch.size(0) 78 | rois_per_image = labels_batch.size(1) 79 | clss = labels_batch 80 | bbox_targets = bbox_target_data.new(batch_size, rois_per_image, 4).zero_() 81 | bbox_inside_weights = bbox_target_data.new(bbox_targets.size()).zero_() 82 | 83 | for b in range(batch_size): 84 | # assert clss[b].sum() > 0 85 | if clss[b].sum() == 0: 86 | continue 87 | inds = torch.nonzero(clss[b] > 0).view(-1) 88 | for i in range(inds.numel()): 89 | ind = inds[i] 90 | bbox_targets[b, ind, :] = bbox_target_data[b, ind, :] 91 | bbox_inside_weights[b, ind, :] = self.BBOX_INSIDE_WEIGHTS 92 | 93 | return bbox_targets, bbox_inside_weights 94 | 95 | 96 | def _compute_targets_pytorch(self, ex_rois, gt_rois): 97 | """Compute bounding-box regression targets for an image.""" 98 | 99 | assert ex_rois.size(1) == gt_rois.size(1) 100 | assert ex_rois.size(2) == 4 101 | assert gt_rois.size(2) == 4 102 | 103 | batch_size = ex_rois.size(0) 104 | rois_per_image = ex_rois.size(1) 105 | 106 | targets = bbox_transform_batch(ex_rois, gt_rois) 107 | 108 | if cfg.TRAIN.BBOX_NORMALIZE_TARGETS_PRECOMPUTED: 109 | # Optionally normalize targets by a precomputed mean and stdev 110 | targets = ((targets - self.BBOX_NORMALIZE_MEANS.expand_as(targets)) 111 | / self.BBOX_NORMALIZE_STDS.expand_as(targets)) 112 | 113 | return targets 114 | 115 | 116 | def _sample_rois_pytorch(self, all_rois, gt_boxes, fg_rois_per_image, rois_per_image, num_classes): 117 | """Generate a random sample of RoIs comprising foreground and background 118 | examples. 119 | """ 120 | # overlaps: (rois x gt_boxes) 121 | 122 | overlaps = bbox_overlaps_batch(all_rois, gt_boxes) 123 | 124 | max_overlaps, gt_assignment = torch.max(overlaps, 2) 125 | 126 | batch_size = overlaps.size(0) 127 | num_proposal = overlaps.size(1) 128 | num_boxes_per_img = overlaps.size(2) 129 | 130 | offset = torch.arange(0, batch_size)*gt_boxes.size(1) 131 | offset = offset.view(-1, 1).type_as(gt_assignment) + gt_assignment 132 | labels = gt_boxes[:, :, 4].contiguous().view(-1)[(offset.view(-1),)].view(batch_size, -1) 133 | #labels = gt_boxes[:,:,4].contiguous().view(-1).index((offset.view(-1),)).view(batch_size, -1) 134 | 135 | labels_batch = labels.new(batch_size, rois_per_image).zero_() 136 | rois_batch = all_rois.new(batch_size, rois_per_image, 5).zero_() 137 | gt_rois_batch = all_rois.new(batch_size, rois_per_image, 5).zero_() 138 | # Guard against the case when an image has fewer than max_fg_rois_per_image 139 | # foreground RoIs 140 | for i in range(batch_size): 141 | 142 | fg_inds = torch.nonzero(max_overlaps[i] >= cfg.TRAIN.FG_THRESH).view(-1) 143 | fg_num_rois = fg_inds.numel() 144 | 145 | # Select background RoIs as those within [BG_THRESH_LO, BG_THRESH_HI) 146 | bg_inds = torch.nonzero((max_overlaps[i] < cfg.TRAIN.BG_THRESH_HI) & 147 | (max_overlaps[i] >= cfg.TRAIN.BG_THRESH_LO)).view(-1) 148 | bg_num_rois = bg_inds.numel() 149 | 150 | if fg_num_rois > 0 and bg_num_rois > 0: 151 | # sampling fg 152 | fg_rois_per_this_image = min(fg_rois_per_image, fg_num_rois) 153 | 154 | # torch.randperm seems has a bug on multi-gpu setting that cause the segfault. 155 | # See https://github.com/pytorch/pytorch/issues/1868 for more details. 156 | # use numpy instead. 157 | #rand_num = torch.randperm(fg_num_rois).long().cuda() 158 | rand_num = torch.from_numpy(np.random.permutation(fg_num_rois)).type_as(gt_boxes).long() 159 | fg_inds = fg_inds[rand_num[:fg_rois_per_this_image]] 160 | 161 | # sampling bg 162 | bg_rois_per_this_image = rois_per_image - fg_rois_per_this_image 163 | 164 | # Seems torch.rand has a bug, it will generate very large number and make an error. 165 | # We use numpy rand instead. 166 | #rand_num = (torch.rand(bg_rois_per_this_image) * bg_num_rois).long().cuda() 167 | rand_num = np.floor(np.random.rand(bg_rois_per_this_image) * bg_num_rois) 168 | rand_num = torch.from_numpy(rand_num).type_as(gt_boxes).long() 169 | bg_inds = bg_inds[rand_num] 170 | 171 | elif fg_num_rois > 0 and bg_num_rois == 0: 172 | # sampling fg 173 | #rand_num = torch.floor(torch.rand(rois_per_image) * fg_num_rois).long().cuda() 174 | rand_num = np.floor(np.random.rand(rois_per_image) * fg_num_rois) 175 | rand_num = torch.from_numpy(rand_num).type_as(gt_boxes).long() 176 | fg_inds = fg_inds[rand_num] 177 | fg_rois_per_this_image = rois_per_image 178 | bg_rois_per_this_image = 0 179 | elif bg_num_rois > 0 and fg_num_rois == 0: 180 | # sampling bg 181 | #rand_num = torch.floor(torch.rand(rois_per_image) * bg_num_rois).long().cuda() 182 | rand_num = np.floor(np.random.rand(rois_per_image) * bg_num_rois) 183 | rand_num = torch.from_numpy(rand_num).type_as(gt_boxes).long() 184 | 185 | bg_inds = bg_inds[rand_num] 186 | bg_rois_per_this_image = rois_per_image 187 | fg_rois_per_this_image = 0 188 | else: 189 | raise ValueError("bg_num_rois = 0 and fg_num_rois = 0, this should not happen!") 190 | 191 | # The indices that we're selecting (both fg and bg) 192 | keep_inds = torch.cat([fg_inds, bg_inds], 0) 193 | 194 | # Select sampled values from various arrays: 195 | labels_batch[i].copy_(labels[i][keep_inds]) 196 | 197 | # Clamp labels for the background RoIs to 0 198 | if fg_rois_per_this_image < rois_per_image: 199 | labels_batch[i][fg_rois_per_this_image:] = 0 200 | 201 | rois_batch[i] = all_rois[i][keep_inds] 202 | rois_batch[i,:,0] = i 203 | 204 | gt_rois_batch[i] = gt_boxes[i][gt_assignment[i][keep_inds]] 205 | 206 | bbox_target_data = self._compute_targets_pytorch( 207 | rois_batch[:,:,1:5], gt_rois_batch[:,:,:4]) 208 | 209 | bbox_targets, bbox_inside_weights = \ 210 | self._get_bbox_regression_labels_pytorch(bbox_target_data, labels_batch, num_classes) 211 | 212 | return labels_batch, rois_batch, bbox_targets, bbox_inside_weights 213 | -------------------------------------------------------------------------------- /lib/model/rpn/rpn.py: -------------------------------------------------------------------------------- 1 | from __future__ import absolute_import 2 | import torch 3 | import torch.nn as nn 4 | import torch.nn.functional as F 5 | from torch.autograd import Variable 6 | 7 | from model.utils.config import cfg 8 | from .proposal_layer import _ProposalLayer 9 | from .anchor_target_layer import _AnchorTargetLayer 10 | from model.utils.net_utils import _smooth_l1_loss 11 | 12 | import numpy as np 13 | import math 14 | import pdb 15 | import time 16 | 17 | class _RPN(nn.Module): 18 | """ region proposal network """ 19 | def __init__(self, din): 20 | super(_RPN, self).__init__() 21 | 22 | self.din = din # get depth of input feature map, e.g., 512 23 | self.anchor_scales = cfg.ANCHOR_SCALES 24 | self.anchor_ratios = cfg.ANCHOR_RATIOS 25 | self.feat_stride = cfg.FEAT_STRIDE[0] 26 | 27 | # define the convrelu layers processing input feature map 28 | self.RPN_Conv = nn.Conv2d(self.din, 512, 3, 1, 1, bias=True) 29 | 30 | # define bg/fg classifcation score layer 31 | self.nc_score_out = len(self.anchor_scales) * len(self.anchor_ratios) * 2 # 2(bg/fg) * 9 (anchors) 32 | self.RPN_cls_score = nn.Conv2d(512, self.nc_score_out, 1, 1, 0) 33 | 34 | # define anchor box offset prediction layer 35 | self.nc_bbox_out = len(self.anchor_scales) * len(self.anchor_ratios) * 4 # 4(coords) * 9 (anchors) 36 | self.RPN_bbox_pred = nn.Conv2d(512, self.nc_bbox_out, 1, 1, 0) 37 | 38 | # define proposal layer 39 | self.RPN_proposal = _ProposalLayer(self.feat_stride, self.anchor_scales, self.anchor_ratios) 40 | 41 | # define anchor target layer 42 | self.RPN_anchor_target = _AnchorTargetLayer(self.feat_stride, self.anchor_scales, self.anchor_ratios) 43 | 44 | self.rpn_loss_cls = 0 45 | self.rpn_loss_box = 0 46 | 47 | @staticmethod 48 | def reshape(x, d): 49 | input_shape = x.size() 50 | x = x.view( 51 | input_shape[0], 52 | int(d), 53 | int(float(input_shape[1] * input_shape[2]) / float(d)), 54 | input_shape[3] 55 | ) 56 | return x 57 | 58 | def forward(self, base_feat, im_info, gt_boxes, num_boxes,target=False): 59 | 60 | batch_size = base_feat.size(0) 61 | 62 | # return feature map after convrelu layer 63 | rpn_conv1 = F.relu(self.RPN_Conv(base_feat), inplace=True) 64 | # get rpn classification score 65 | rpn_cls_score = self.RPN_cls_score(rpn_conv1) 66 | 67 | rpn_cls_score_reshape = self.reshape(rpn_cls_score, 2) 68 | rpn_cls_prob_reshape = F.softmax(rpn_cls_score_reshape, 1) 69 | rpn_cls_prob = self.reshape(rpn_cls_prob_reshape, self.nc_score_out) 70 | 71 | # get rpn offsets to the anchor boxes 72 | rpn_bbox_pred = self.RPN_bbox_pred(rpn_conv1) 73 | 74 | # proposal layer 75 | cfg_key = 'TRAIN' if self.training else 'TEST' 76 | 77 | rois = self.RPN_proposal((rpn_cls_prob.data, rpn_bbox_pred.data, 78 | im_info, cfg_key),target=target) 79 | 80 | self.rpn_loss_cls = 0 81 | self.rpn_loss_box = 0 82 | 83 | # generating training labels and build the rpn loss 84 | if self.training: 85 | assert gt_boxes is not None 86 | 87 | rpn_data = self.RPN_anchor_target((rpn_cls_score.data, gt_boxes, im_info, num_boxes)) 88 | 89 | # compute classification loss 90 | rpn_cls_score = rpn_cls_score_reshape.permute(0, 2, 3, 1).contiguous().view(batch_size, -1, 2) 91 | rpn_label = rpn_data[0].view(batch_size, -1) 92 | 93 | rpn_keep = Variable(rpn_label.view(-1).ne(-1).nonzero().view(-1)) 94 | rpn_cls_score = torch.index_select(rpn_cls_score.view(-1,2), 0, rpn_keep) 95 | rpn_label = torch.index_select(rpn_label.view(-1), 0, rpn_keep.data) 96 | rpn_label = Variable(rpn_label.long()) 97 | self.rpn_loss_cls = F.cross_entropy(rpn_cls_score, rpn_label) 98 | fg_cnt = torch.sum(rpn_label.data.ne(0)) 99 | 100 | rpn_bbox_targets, rpn_bbox_inside_weights, rpn_bbox_outside_weights = rpn_data[1:] 101 | 102 | # compute bbox regression loss 103 | rpn_bbox_inside_weights = Variable(rpn_bbox_inside_weights) 104 | rpn_bbox_outside_weights = Variable(rpn_bbox_outside_weights) 105 | rpn_bbox_targets = Variable(rpn_bbox_targets) 106 | 107 | self.rpn_loss_box = _smooth_l1_loss(rpn_bbox_pred, rpn_bbox_targets, rpn_bbox_inside_weights, 108 | rpn_bbox_outside_weights, sigma=3, dim=[1,2,3]) 109 | 110 | return rois, self.rpn_loss_cls, self.rpn_loss_box 111 | -------------------------------------------------------------------------------- /lib/model/utils/.gitignore: -------------------------------------------------------------------------------- 1 | *.c 2 | *.cpp 3 | *.so 4 | -------------------------------------------------------------------------------- /lib/model/utils/__init__.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/basiclab/DA-OD-MEAA-PyTorch/ab590ddacdd06886fdbd56f7c4f59cd36167d100/lib/model/utils/__init__.py -------------------------------------------------------------------------------- /lib/model/utils/bbox.pyx: -------------------------------------------------------------------------------- 1 | # -------------------------------------------------------- 2 | # Fast R-CNN 3 | # Copyright (c) 2015 Microsoft 4 | # Licensed under The MIT License [see LICENSE for details] 5 | # Written by Sergey Karayev 6 | # -------------------------------------------------------- 7 | 8 | cimport cython 9 | import numpy as np 10 | cimport numpy as np 11 | 12 | DTYPE = np.float 13 | ctypedef np.float_t DTYPE_t 14 | 15 | def bbox_overlaps(np.ndarray[DTYPE_t, ndim=2] boxes, 16 | np.ndarray[DTYPE_t, ndim=2] query_boxes): 17 | return bbox_overlaps_c(boxes, query_boxes) 18 | 19 | cdef np.ndarray[DTYPE_t, ndim=2] bbox_overlaps_c( 20 | np.ndarray[DTYPE_t, ndim=2] boxes, 21 | np.ndarray[DTYPE_t, ndim=2] query_boxes): 22 | """ 23 | Parameters 24 | ---------- 25 | boxes: (N, 4) ndarray of float 26 | query_boxes: (K, 4) ndarray of float 27 | Returns 28 | ------- 29 | overlaps: (N, K) ndarray of overlap between boxes and query_boxes 30 | """ 31 | cdef unsigned int N = boxes.shape[0] 32 | cdef unsigned int K = query_boxes.shape[0] 33 | cdef np.ndarray[DTYPE_t, ndim=2] overlaps = np.zeros((N, K), dtype=DTYPE) 34 | cdef DTYPE_t iw, ih, box_area 35 | cdef DTYPE_t ua 36 | cdef unsigned int k, n 37 | for k in range(K): 38 | box_area = ( 39 | (query_boxes[k, 2] - query_boxes[k, 0] + 1) * 40 | (query_boxes[k, 3] - query_boxes[k, 1] + 1) 41 | ) 42 | for n in range(N): 43 | iw = ( 44 | min(boxes[n, 2], query_boxes[k, 2]) - 45 | max(boxes[n, 0], query_boxes[k, 0]) + 1 46 | ) 47 | if iw > 0: 48 | ih = ( 49 | min(boxes[n, 3], query_boxes[k, 3]) - 50 | max(boxes[n, 1], query_boxes[k, 1]) + 1 51 | ) 52 | if ih > 0: 53 | ua = float( 54 | (boxes[n, 2] - boxes[n, 0] + 1) * 55 | (boxes[n, 3] - boxes[n, 1] + 1) + 56 | box_area - iw * ih 57 | ) 58 | overlaps[n, k] = iw * ih / ua 59 | return overlaps 60 | 61 | 62 | def bbox_intersections( 63 | np.ndarray[DTYPE_t, ndim=2] boxes, 64 | np.ndarray[DTYPE_t, ndim=2] query_boxes): 65 | return bbox_intersections_c(boxes, query_boxes) 66 | 67 | 68 | cdef np.ndarray[DTYPE_t, ndim=2] bbox_intersections_c( 69 | np.ndarray[DTYPE_t, ndim=2] boxes, 70 | np.ndarray[DTYPE_t, ndim=2] query_boxes): 71 | """ 72 | For each query box compute the intersection ratio covered by boxes 73 | ---------- 74 | Parameters 75 | ---------- 76 | boxes: (N, 4) ndarray of float 77 | query_boxes: (K, 4) ndarray of float 78 | Returns 79 | ------- 80 | overlaps: (N, K) ndarray of intersec between boxes and query_boxes 81 | """ 82 | cdef unsigned int N = boxes.shape[0] 83 | cdef unsigned int K = query_boxes.shape[0] 84 | cdef np.ndarray[DTYPE_t, ndim=2] intersec = np.zeros((N, K), dtype=DTYPE) 85 | cdef DTYPE_t iw, ih, box_area 86 | cdef DTYPE_t ua 87 | cdef unsigned int k, n 88 | for k in range(K): 89 | box_area = ( 90 | (query_boxes[k, 2] - query_boxes[k, 0] + 1) * 91 | (query_boxes[k, 3] - query_boxes[k, 1] + 1) 92 | ) 93 | for n in range(N): 94 | iw = ( 95 | min(boxes[n, 2], query_boxes[k, 2]) - 96 | max(boxes[n, 0], query_boxes[k, 0]) + 1 97 | ) 98 | if iw > 0: 99 | ih = ( 100 | min(boxes[n, 3], query_boxes[k, 3]) - 101 | max(boxes[n, 1], query_boxes[k, 1]) + 1 102 | ) 103 | if ih > 0: 104 | intersec[n, k] = iw * ih / box_area 105 | return intersec -------------------------------------------------------------------------------- /lib/model/utils/blob.py: -------------------------------------------------------------------------------- 1 | # -------------------------------------------------------- 2 | # Fast R-CNN 3 | # Copyright (c) 2015 Microsoft 4 | # Licensed under The MIT License [see LICENSE for details] 5 | # Written by Ross Girshick 6 | # -------------------------------------------------------- 7 | 8 | """Blob helper functions.""" 9 | 10 | import numpy as np 11 | # from scipy.misc import imread, imresize 12 | import cv2 13 | 14 | try: 15 | xrange # Python 2 16 | except NameError: 17 | xrange = range # Python 3 18 | 19 | 20 | def im_list_to_blob(ims): 21 | """Convert a list of images into a network input. 22 | 23 | Assumes images are already prepared (means subtracted, BGR order, ...). 24 | """ 25 | max_shape = np.array([im.shape for im in ims]).max(axis=0) 26 | num_images = len(ims) 27 | blob = np.zeros((num_images, max_shape[0], max_shape[1], 3), 28 | dtype=np.float32) 29 | for i in xrange(num_images): 30 | im = ims[i] 31 | blob[i, 0:im.shape[0], 0:im.shape[1], :] = im 32 | 33 | return blob 34 | 35 | def prep_im_for_blob(im, pixel_means, target_size, max_size): 36 | """Mean subtract and scale an image for use in a blob.""" 37 | 38 | im = im.astype(np.float32, copy=False) 39 | im -= pixel_means 40 | # im = im[:, :, ::-1] 41 | im_shape = im.shape 42 | im_size_min = np.min(im_shape[0:2]) 43 | im_size_max = np.max(im_shape[0:2]) 44 | im_scale = float(target_size) / float(im_size_min) 45 | # Prevent the biggest axis from being more than MAX_SIZE 46 | #if np.round(im_scale * im_size_max) > max_size: 47 | # im_scale = float(max_size) / float(im_size_max) 48 | # im = imresize(im, im_scale) 49 | im = cv2.resize(im, None, None, fx=im_scale, fy=im_scale, 50 | interpolation=cv2.INTER_LINEAR) 51 | 52 | return im, im_scale 53 | -------------------------------------------------------------------------------- /lib/pycocotools/UPSTREAM_REV: -------------------------------------------------------------------------------- 1 | https://github.com/pdollar/coco/commit/3ac47c77ebd5a1ed4254a98b7fbf2ef4765a3574 2 | -------------------------------------------------------------------------------- /lib/pycocotools/__init__.py: -------------------------------------------------------------------------------- 1 | __author__ = 'tylin' 2 | -------------------------------------------------------------------------------- /lib/pycocotools/license.txt: -------------------------------------------------------------------------------- 1 | Copyright (c) 2014, Piotr Dollar and Tsung-Yi Lin 2 | All rights reserved. 3 | 4 | Redistribution and use in source and binary forms, with or without 5 | modification, are permitted provided that the following conditions are met: 6 | 7 | 1. Redistributions of source code must retain the above copyright notice, this 8 | list of conditions and the following disclaimer. 9 | 2. Redistributions in binary form must reproduce the above copyright notice, 10 | this list of conditions and the following disclaimer in the documentation 11 | and/or other materials provided with the distribution. 12 | 13 | THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND 14 | ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED 15 | WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE 16 | DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR 17 | ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES 18 | (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; 19 | LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND 20 | ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 21 | (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS 22 | SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 23 | 24 | The views and conclusions contained in the software and documentation are those 25 | of the authors and should not be interpreted as representing official policies, 26 | either expressed or implied, of the FreeBSD Project. 27 | -------------------------------------------------------------------------------- /lib/pycocotools/mask.py: -------------------------------------------------------------------------------- 1 | __author__ = 'tsungyi' 2 | 3 | from . import _mask 4 | 5 | # Interface for manipulating masks stored in RLE format. 6 | # 7 | # RLE is a simple yet efficient format for storing binary masks. RLE 8 | # first divides a vector (or vectorized image) into a series of piecewise 9 | # constant regions and then for each piece simply stores the length of 10 | # that piece. For example, given M=[0 0 1 1 1 0 1] the RLE counts would 11 | # be [2 3 1 1], or for M=[1 1 1 1 1 1 0] the counts would be [0 6 1] 12 | # (note that the odd counts are always the numbers of zeros). Instead of 13 | # storing the counts directly, additional compression is achieved with a 14 | # variable bitrate representation based on a common scheme called LEB128. 15 | # 16 | # Compression is greatest given large piecewise constant regions. 17 | # Specifically, the size of the RLE is proportional to the number of 18 | # *boundaries* in M (or for an image the number of boundaries in the y 19 | # direction). Assuming fairly simple shapes, the RLE representation is 20 | # O(sqrt(n)) where n is number of pixels in the object. Hence space usage 21 | # is substantially lower, especially for large simple objects (large n). 22 | # 23 | # Many common operations on masks can be computed directly using the RLE 24 | # (without need for decoding). This includes computations such as area, 25 | # union, intersection, etc. All of these operations are linear in the 26 | # size of the RLE, in other words they are O(sqrt(n)) where n is the area 27 | # of the object. Computing these operations on the original mask is O(n). 28 | # Thus, using the RLE can result in substantial computational savings. 29 | # 30 | # The following API functions are defined: 31 | # encode - Encode binary masks using RLE. 32 | # decode - Decode binary masks encoded via RLE. 33 | # merge - Compute union or intersection of encoded masks. 34 | # iou - Compute intersection over union between masks. 35 | # area - Compute area of encoded masks. 36 | # toBbox - Get bounding boxes surrounding encoded masks. 37 | # frPyObjects - Convert polygon, bbox, and uncompressed RLE to encoded RLE mask. 38 | # 39 | # Usage: 40 | # Rs = encode( masks ) 41 | # masks = decode( Rs ) 42 | # R = merge( Rs, intersect=false ) 43 | # o = iou( dt, gt, iscrowd ) 44 | # a = area( Rs ) 45 | # bbs = toBbox( Rs ) 46 | # Rs = frPyObjects( [pyObjects], h, w ) 47 | # 48 | # In the API the following formats are used: 49 | # Rs - [dict] Run-length encoding of binary masks 50 | # R - dict Run-length encoding of binary mask 51 | # masks - [hxwxn] Binary mask(s) (must have type np.ndarray(dtype=uint8) in column-major order) 52 | # iscrowd - [nx1] list of np.ndarray. 1 indicates corresponding gt image has crowd region to ignore 53 | # bbs - [nx4] Bounding box(es) stored as [x y w h] 54 | # poly - Polygon stored as [[x1 y1 x2 y2...],[x1 y1 ...],...] (2D list) 55 | # dt,gt - May be either bounding boxes or encoded masks 56 | # Both poly and bbs are 0-indexed (bbox=[0 0 1 1] encloses first pixel). 57 | # 58 | # Finally, a note about the intersection over union (iou) computation. 59 | # The standard iou of a ground truth (gt) and detected (dt) object is 60 | # iou(gt,dt) = area(intersect(gt,dt)) / area(union(gt,dt)) 61 | # For "crowd" regions, we use a modified criteria. If a gt object is 62 | # marked as "iscrowd", we allow a dt to match any subregion of the gt. 63 | # Choosing gt' in the crowd gt that best matches the dt can be done using 64 | # gt'=intersect(dt,gt). Since by definition union(gt',dt)=dt, computing 65 | # iou(gt,dt,iscrowd) = iou(gt',dt) = area(intersect(gt,dt)) / area(dt) 66 | # For crowd gt regions we use this modified criteria above for the iou. 67 | # 68 | # To compile run "python setup.py build_ext --inplace" 69 | # Please do not contact us for help with compiling. 70 | # 71 | # Microsoft COCO Toolbox. version 2.0 72 | # Data, paper, and tutorials available at: http://mscoco.org/ 73 | # Code written by Piotr Dollar and Tsung-Yi Lin, 2015. 74 | # Licensed under the Simplified BSD License [see coco/license.txt] 75 | 76 | encode = _mask.encode 77 | decode = _mask.decode 78 | iou = _mask.iou 79 | merge = _mask.merge 80 | area = _mask.area 81 | toBbox = _mask.toBbox 82 | frPyObjects = _mask.frPyObjects -------------------------------------------------------------------------------- /lib/pycocotools/maskApi.c: -------------------------------------------------------------------------------- 1 | /************************************************************************** 2 | * Microsoft COCO Toolbox. version 2.0 3 | * Data, paper, and tutorials available at: http://mscoco.org/ 4 | * Code written by Piotr Dollar and Tsung-Yi Lin, 2015. 5 | * Licensed under the Simplified BSD License [see coco/license.txt] 6 | **************************************************************************/ 7 | #include "maskApi.h" 8 | #include 9 | #include 10 | 11 | uint umin( uint a, uint b ) { return (ab) ? a : b; } 13 | 14 | void rleInit( RLE *R, siz h, siz w, siz m, uint *cnts ) { 15 | R->h=h; R->w=w; R->m=m; R->cnts=(m==0)?0:malloc(sizeof(uint)*m); 16 | if(cnts) for(siz j=0; jcnts[j]=cnts[j]; 17 | } 18 | 19 | void rleFree( RLE *R ) { 20 | free(R->cnts); R->cnts=0; 21 | } 22 | 23 | void rlesInit( RLE **R, siz n ) { 24 | *R = (RLE*) malloc(sizeof(RLE)*n); 25 | for(siz i=0; i0 ) { 61 | c=umin(ca,cb); cc+=c; ct=0; 62 | ca-=c; if(!ca && a0) { 83 | crowd=iscrowd!=NULL && iscrowd[g]; 84 | if(dt[d].h!=gt[g].h || dt[d].w!=gt[g].w) { o[g*m+d]=-1; continue; } 85 | siz ka, kb, a, b; uint c, ca, cb, ct, i, u; bool va, vb; 86 | ca=dt[d].cnts[0]; ka=dt[d].m; va=vb=0; 87 | cb=gt[g].cnts[0]; kb=gt[g].m; a=b=1; i=u=0; ct=1; 88 | while( ct>0 ) { 89 | c=umin(ca,cb); if(va||vb) { u+=c; if(va&&vb) i+=c; } ct=0; 90 | ca-=c; if(!ca && ad?1:c=dy && xs>xe) || (dxye); 151 | if(flip) { t=xs; xs=xe; xe=t; t=ys; ys=ye; ye=t; } 152 | s = dx>=dy ? (double)(ye-ys)/dx : (double)(xe-xs)/dy; 153 | if(dx>=dy) for( int d=0; d<=dx; d++ ) { 154 | t=flip?dx-d:d; u[m]=t+xs; v[m]=(int)(ys+s*t+.5); m++; 155 | } else for( int d=0; d<=dy; d++ ) { 156 | t=flip?dy-d:d; v[m]=t+ys; u[m]=(int)(xs+s*t+.5); m++; 157 | } 158 | } 159 | // get points along y-boundary and downsample 160 | free(x); free(y); k=m; m=0; double xd, yd; 161 | x=malloc(sizeof(int)*k); y=malloc(sizeof(int)*k); 162 | for( j=1; jw-1 ) continue; 165 | yd=(double)(v[j]h) yd=h; yd=ceil(yd); 167 | x[m]=(int) xd; y[m]=(int) yd; m++; 168 | } 169 | // compute rle encoding given y-boundary points 170 | k=m; a=malloc(sizeof(uint)*(k+1)); 171 | for( j=0; j0) b[m++]=a[j++]; else { 177 | j++; if(jm, p=0; long x; bool more; 184 | char *s=malloc(sizeof(char)*m*6); 185 | for( i=0; icnts[i]; if(i>2) x-=(long) R->cnts[i-2]; more=1; 187 | while( more ) { 188 | char c=x & 0x1f; x >>= 5; more=(c & 0x10) ? x!=-1 : x!=0; 189 | if(more) c |= 0x20; c+=48; s[p++]=c; 190 | } 191 | } 192 | s[p]=0; return s; 193 | } 194 | 195 | void rleFrString( RLE *R, char *s, siz h, siz w ) { 196 | siz m=0, p=0, k; long x; bool more; uint *cnts; 197 | while( s[m] ) m++; cnts=malloc(sizeof(uint)*m); m=0; 198 | while( s[p] ) { 199 | x=0; k=0; more=1; 200 | while( more ) { 201 | char c=s[p]-48; x |= (c & 0x1f) << 5*k; 202 | more = c & 0x20; p++; k++; 203 | if(!more && (c & 0x10)) x |= -1 << 5*k; 204 | } 205 | if(m>2) x+=(long) cnts[m-2]; cnts[m++]=(uint) x; 206 | } 207 | rleInit(R,h,w,m,cnts); free(cnts); 208 | } 209 | -------------------------------------------------------------------------------- /lib/pycocotools/maskApi.h: -------------------------------------------------------------------------------- 1 | /************************************************************************** 2 | * Microsoft COCO Toolbox. version 2.0 3 | * Data, paper, and tutorials available at: http://mscoco.org/ 4 | * Code written by Piotr Dollar and Tsung-Yi Lin, 2015. 5 | * Licensed under the Simplified BSD License [see coco/license.txt] 6 | **************************************************************************/ 7 | #pragma once 8 | #include 9 | 10 | typedef unsigned int uint; 11 | typedef unsigned long siz; 12 | typedef unsigned char byte; 13 | typedef double* BB; 14 | typedef struct { siz h, w, m; uint *cnts; } RLE; 15 | 16 | // Initialize/destroy RLE. 17 | void rleInit( RLE *R, siz h, siz w, siz m, uint *cnts ); 18 | void rleFree( RLE *R ); 19 | 20 | // Initialize/destroy RLE array. 21 | void rlesInit( RLE **R, siz n ); 22 | void rlesFree( RLE **R, siz n ); 23 | 24 | // Encode binary masks using RLE. 25 | void rleEncode( RLE *R, const byte *mask, siz h, siz w, siz n ); 26 | 27 | // Decode binary masks encoded via RLE. 28 | void rleDecode( const RLE *R, byte *mask, siz n ); 29 | 30 | // Compute union or intersection of encoded masks. 31 | void rleMerge( const RLE *R, RLE *M, siz n, bool intersect ); 32 | 33 | // Compute area of encoded masks. 34 | void rleArea( const RLE *R, siz n, uint *a ); 35 | 36 | // Compute intersection over union between masks. 37 | void rleIou( RLE *dt, RLE *gt, siz m, siz n, byte *iscrowd, double *o ); 38 | 39 | // Compute intersection over union between bounding boxes. 40 | void bbIou( BB dt, BB gt, siz m, siz n, byte *iscrowd, double *o ); 41 | 42 | // Get bounding boxes surrounding encoded masks. 43 | void rleToBbox( const RLE *R, BB bb, siz n ); 44 | 45 | // Convert bounding boxes to encoded masks. 46 | void rleFrBbox( RLE *R, const BB bb, siz h, siz w, siz n ); 47 | 48 | // Convert polygon to encoded mask. 49 | void rleFrPoly( RLE *R, const double *xy, siz k, siz h, siz w ); 50 | 51 | // Get compressed string representation of encoded mask. 52 | char* rleToString( const RLE *R ); 53 | 54 | // Convert from compressed string representation of encoded mask. 55 | void rleFrString( RLE *R, char *s, siz h, siz w ); 56 | -------------------------------------------------------------------------------- /lib/roi_data_layer/__init__.py: -------------------------------------------------------------------------------- 1 | # -------------------------------------------------------- 2 | # Fast R-CNN 3 | # Copyright (c) 2015 Microsoft 4 | # Licensed under The MIT License [see LICENSE for details] 5 | # Written by Ross Girshick 6 | # -------------------------------------------------------- 7 | -------------------------------------------------------------------------------- /lib/roi_data_layer/minibatch.py: -------------------------------------------------------------------------------- 1 | # -------------------------------------------------------- 2 | # Fast R-CNN 3 | # Copyright (c) 2015 Microsoft 4 | # Licensed under The MIT License [see LICENSE for details] 5 | # Written by Ross Girshick and Xinlei Chen 6 | # -------------------------------------------------------- 7 | 8 | """Compute minibatch blobs for training a Fast R-CNN network.""" 9 | from __future__ import absolute_import 10 | from __future__ import division 11 | from __future__ import print_function 12 | 13 | import numpy as np 14 | import numpy.random as npr 15 | from scipy.misc import imread 16 | from model.utils.config import cfg 17 | from model.utils.blob import prep_im_for_blob, im_list_to_blob 18 | import pdb 19 | def get_minibatch(roidb, num_classes,seg_return=False): 20 | """Given a roidb, construct a minibatch sampled from it.""" 21 | num_images = len(roidb) 22 | # Sample random scales to use for each image in this batch 23 | random_scale_inds = npr.randint(0, high=len(cfg.TRAIN.SCALES), 24 | size=num_images) 25 | assert(cfg.TRAIN.BATCH_SIZE % num_images == 0), \ 26 | 'num_images ({}) must divide BATCH_SIZE ({})'. \ 27 | format(num_images, cfg.TRAIN.BATCH_SIZE) 28 | 29 | # Get the input image blob, formatted for caffe 30 | im_blob, im_scales = _get_image_blob(roidb, random_scale_inds) 31 | 32 | blobs = {'data': im_blob} 33 | 34 | assert len(im_scales) == 1, "Single batch only" 35 | assert len(roidb) == 1, "Single batch only" 36 | 37 | # gt boxes: (x1, y1, x2, y2, cls) 38 | if cfg.TRAIN.USE_ALL_GT: 39 | # Include all ground truth boxes 40 | gt_inds = np.where(roidb[0]['gt_classes'] != 0)[0] 41 | else: 42 | # For the COCO ground truth boxes, exclude the ones that are ''iscrowd'' 43 | gt_inds = np.where((roidb[0]['gt_classes'] != 0) & np.all(roidb[0]['gt_overlaps'].toarray() > -1.0, axis=1))[0] 44 | gt_boxes = np.empty((len(gt_inds), 5), dtype=np.float32) 45 | gt_boxes[:, 0:4] = roidb[0]['boxes'][gt_inds, :] * im_scales[0] 46 | gt_boxes[:, 4] = roidb[0]['gt_classes'][gt_inds] 47 | blobs['gt_boxes'] = gt_boxes 48 | blobs['im_info'] = np.array( 49 | [[im_blob.shape[1], im_blob.shape[2], im_scales[0]]], 50 | dtype=np.float32) 51 | if seg_return: 52 | blobs['seg_map'] = roidb[0]['seg_map'] 53 | blobs['img_id'] = roidb[0]['img_id'] 54 | blobs['path'] = roidb[0]['image'] 55 | 56 | return blobs 57 | 58 | def _get_image_blob(roidb, scale_inds): 59 | """Builds an input blob from the images in the roidb at the specified 60 | scales. 61 | """ 62 | num_images = len(roidb) 63 | 64 | processed_ims = [] 65 | im_scales = [] 66 | for i in range(num_images): 67 | #im = cv2.imread(roidb[i]['image']) 68 | im = imread(roidb[i]['image']) 69 | 70 | if len(im.shape) == 2: 71 | im = im[:,:,np.newaxis] 72 | im = np.concatenate((im,im,im), axis=2) 73 | # flip the channel, since the original one using cv2 74 | # rgb -> bgr 75 | im = im[:,:,::-1] 76 | 77 | if roidb[i]['flipped']: 78 | im = im[:, ::-1, :] 79 | target_size = cfg.TRAIN.SCALES[scale_inds[i]] 80 | im, im_scale = prep_im_for_blob(im, cfg.PIXEL_MEANS, target_size, 81 | cfg.TRAIN.MAX_SIZE) 82 | im_scales.append(im_scale) 83 | processed_ims.append(im) 84 | 85 | # Create a blob to hold the input images 86 | blob = im_list_to_blob(processed_ims) 87 | 88 | return blob, im_scales 89 | -------------------------------------------------------------------------------- /lib/roi_data_layer/roibatchLoader.py: -------------------------------------------------------------------------------- 1 | 2 | """The data layer used during training to train a Fast R-CNN network. 3 | """ 4 | 5 | from __future__ import absolute_import 6 | from __future__ import division 7 | from __future__ import print_function 8 | 9 | import torch.utils.data as data 10 | from PIL import Image 11 | import torch 12 | 13 | from model.utils.config import cfg 14 | from roi_data_layer.minibatch import get_minibatch, get_minibatch 15 | from model.rpn.bbox_transform import bbox_transform_inv, clip_boxes 16 | 17 | import numpy as np 18 | import random 19 | import time 20 | import pdb 21 | class roibatchLoader(data.Dataset): 22 | def __init__(self, roidb, ratio_list, ratio_index, batch_size, num_classes, training=True, normalize=None,seg_return=False,path_return=False): 23 | self._roidb = roidb 24 | self._num_classes = num_classes 25 | # we make the height of image consistent to trim_height, trim_width 26 | self.trim_height = cfg.TRAIN.TRIM_HEIGHT 27 | self.trim_width = cfg.TRAIN.TRIM_WIDTH 28 | self.max_num_box = cfg.MAX_NUM_GT_BOXES 29 | self.training = training 30 | self.normalize = normalize 31 | self.ratio_list = ratio_list 32 | self.ratio_index = ratio_index 33 | self.batch_size = batch_size 34 | self.data_size = len(self.ratio_list) 35 | self.seg_return = seg_return 36 | self.path_return = path_return 37 | # given the ratio_list, we want to make the ratio same for each batch. 38 | self.ratio_list_batch = torch.Tensor(self.data_size).zero_() 39 | num_batch = int(np.ceil(len(ratio_index) / batch_size)) 40 | for i in range(num_batch): 41 | left_idx = i*batch_size 42 | right_idx = min((i+1)*batch_size-1, self.data_size-1) 43 | 44 | if ratio_list[right_idx] < 1: 45 | # for ratio < 1, we preserve the leftmost in each batch. 46 | target_ratio = ratio_list[left_idx] 47 | elif ratio_list[left_idx] > 1: 48 | # for ratio > 1, we preserve the rightmost in each batch. 49 | target_ratio = ratio_list[right_idx] 50 | else: 51 | # for ratio cross 1, we make it to be 1. 52 | target_ratio = 1 53 | 54 | self.ratio_list_batch[left_idx:(right_idx+1)] = target_ratio 55 | 56 | 57 | def __getitem__(self, index): 58 | if self.training: 59 | index_ratio = int(self.ratio_index[index]) 60 | else: 61 | index_ratio = index 62 | 63 | # get the anchor index for current sample index 64 | # here we set the anchor index to the last one 65 | # sample in this group 66 | minibatch_db = [self._roidb[index_ratio]] 67 | 68 | blobs = get_minibatch(minibatch_db, self._num_classes,seg_return=self.seg_return) 69 | data = torch.from_numpy(blobs['data']) 70 | im_info = torch.from_numpy(blobs['im_info']) 71 | # we need to random shuffle the bounding box. 72 | data_height, data_width = data.size(1), data.size(2) 73 | if self.training: 74 | np.random.shuffle(blobs['gt_boxes']) 75 | gt_boxes = torch.from_numpy(blobs['gt_boxes']) # correct loaded gt_bbox 76 | # print('\n###### gt_boxes: ', gt_boxes) 77 | 78 | ######################################################## 79 | # padding the input image to fixed size for each group # 80 | ######################################################## 81 | 82 | # NOTE1: need to cope with the case where a group cover both conditions. (done) 83 | # NOTE2: need to consider the situation for the tail samples. (no worry) 84 | # NOTE3: need to implement a parallel data loader. (no worry) 85 | # get the index range 86 | 87 | # if the image need to crop, crop to the target size. 88 | ratio = self.ratio_list_batch[index] 89 | 90 | if self._roidb[index_ratio]['need_crop']: 91 | if ratio < 1: 92 | # this means that data_width << data_height, we need to crop the 93 | # data_height 94 | min_y = int(torch.min(gt_boxes[:,1])) 95 | max_y = int(torch.max(gt_boxes[:,3])) 96 | 97 | trim_size = int(np.floor(data_width / ratio)) 98 | if trim_size > data_height: 99 | trim_size = data_height 100 | box_region = max_y - min_y + 1 101 | if min_y == 0: 102 | y_s = 0 103 | else: 104 | if (box_region-trim_size) < 0: 105 | y_s_min = max(max_y-trim_size, 0) 106 | y_s_max = min(min_y, data_height-trim_size) 107 | if y_s_min == y_s_max: 108 | y_s = y_s_min 109 | else: 110 | y_s = np.random.choice(range(y_s_min, y_s_max)) 111 | else: 112 | y_s_add = int((box_region-trim_size)/2) 113 | if y_s_add == 0: 114 | y_s = min_y 115 | else: 116 | y_s = np.random.choice(range(min_y, min_y+y_s_add)) 117 | # crop the image 118 | data = data[:, y_s:(y_s + trim_size), :, :] 119 | 120 | # shift y coordiante of gt_boxes 121 | gt_boxes[:, 1] = gt_boxes[:, 1] - float(y_s) 122 | gt_boxes[:, 3] = gt_boxes[:, 3] - float(y_s) 123 | 124 | # update gt bounding box according the trip 125 | gt_boxes[:, 1].clamp_(0, trim_size - 1) 126 | gt_boxes[:, 3].clamp_(0, trim_size - 1) 127 | 128 | else: 129 | # this means that data_width >> data_height, we need to crop the 130 | # data_width 131 | min_x = int(torch.min(gt_boxes[:,0])) 132 | max_x = int(torch.max(gt_boxes[:,2])) 133 | trim_size = int(np.ceil(data_height * ratio)) 134 | if trim_size > data_width: 135 | trim_size = data_width 136 | box_region = max_x - min_x + 1 137 | if min_x == 0: 138 | x_s = 0 139 | else: 140 | if (box_region-trim_size) < 0: 141 | x_s_min = max(max_x-trim_size, 0) 142 | x_s_max = min(min_x, data_width-trim_size) 143 | if x_s_min == x_s_max: 144 | x_s = x_s_min 145 | else: 146 | x_s = np.random.choice(range(x_s_min, x_s_max)) 147 | else: 148 | x_s_add = int((box_region-trim_size)/2) 149 | if x_s_add == 0: 150 | x_s = min_x 151 | else: 152 | x_s = np.random.choice(range(min_x, min_x+x_s_add)) 153 | # crop the image 154 | data = data[:, :, x_s:(x_s + trim_size), :] 155 | 156 | # shift x coordiante of gt_boxes 157 | gt_boxes[:, 0] = gt_boxes[:, 0] - float(x_s) 158 | gt_boxes[:, 2] = gt_boxes[:, 2] - float(x_s) 159 | # update gt bounding box according the trip 160 | gt_boxes[:, 0].clamp_(0, trim_size - 1) 161 | gt_boxes[:, 2].clamp_(0, trim_size - 1) 162 | 163 | # based on the ratio, padding the image. 164 | if ratio < 1: 165 | # this means that data_width < data_height 166 | trim_size = int(np.floor(data_width / ratio)) 167 | 168 | padding_data = torch.FloatTensor(int(np.ceil(data_width / ratio)), \ 169 | data_width, 3).zero_() 170 | 171 | padding_data[:data_height, :, :] = data[0] 172 | # update im_info 173 | im_info[0, 0] = padding_data.size(0) 174 | # print("height %d %d \n" %(index, anchor_idx)) 175 | elif ratio > 1: 176 | # this means that data_width > data_height 177 | # if the image need to crop. 178 | padding_data = torch.FloatTensor(data_height, \ 179 | int(np.ceil(data_height * ratio)), 3).zero_() 180 | padding_data[:, :data_width, :] = data[0] 181 | im_info[0, 1] = padding_data.size(1) 182 | else: 183 | trim_size = min(data_height, data_width) 184 | padding_data = torch.FloatTensor(trim_size, trim_size, 3).zero_() 185 | padding_data = data[0][:trim_size, :trim_size, :] 186 | # gt_boxes.clamp_(0, trim_size) 187 | gt_boxes[:, :4].clamp_(0, trim_size) 188 | im_info[0, 0] = trim_size 189 | im_info[0, 1] = trim_size 190 | 191 | 192 | # check the bounding box: 193 | not_keep = (gt_boxes[:,0] == gt_boxes[:,2]) | (gt_boxes[:,1] == gt_boxes[:,3]) 194 | keep = torch.nonzero(not_keep == 0).view(-1) 195 | 196 | gt_boxes_padding = torch.FloatTensor(self.max_num_box, gt_boxes.size(1)).zero_() 197 | 198 | 199 | if keep.numel() != 0: 200 | gt_boxes = gt_boxes[keep] 201 | num_boxes = min(gt_boxes.size(0), self.max_num_box) 202 | gt_boxes_padding[:num_boxes,:] = gt_boxes[:num_boxes] 203 | else: 204 | num_boxes = 0 205 | 206 | 207 | # print('\n###### gt_boxes: ', gt_boxes_padding) # worked 208 | # permute trim_data to adapt to downstream processing 209 | padding_data = padding_data.permute(2, 0, 1).contiguous() 210 | im_info = im_info.view(3) 211 | if self.seg_return: 212 | seg_map = torch.from_numpy(np.resize(blobs['seg_map'],(data_height,data_width))) 213 | return padding_data, im_info, gt_boxes_padding, num_boxes, seg_map 214 | elif self.path_return: 215 | return padding_data, im_info, gt_boxes_padding, num_boxes, blobs['path'] 216 | else: 217 | return padding_data, im_info, gt_boxes_padding, num_boxes 218 | 219 | 220 | else: # testing 221 | data = data.permute(0, 3, 1, 2).contiguous().view(3, data_height, data_width) 222 | im_info = im_info.view(3) 223 | 224 | gt_boxes = torch.FloatTensor([1,1,1,1,1]) 225 | num_boxes = 0 226 | 227 | return data, im_info, gt_boxes, num_boxes, blobs['path'] 228 | 229 | def __len__(self): 230 | return len(self._roidb) 231 | -------------------------------------------------------------------------------- /lib/roi_data_layer/roidb.py: -------------------------------------------------------------------------------- 1 | """Transform a roidb into a trainable roidb by adding a bunch of metadata.""" 2 | from __future__ import absolute_import 3 | from __future__ import division 4 | from __future__ import print_function 5 | 6 | import datasets 7 | import numpy as np 8 | from model.utils.config import cfg 9 | from datasets.factory import get_imdb 10 | import PIL 11 | import pdb 12 | 13 | def prepare_roidb(imdb): 14 | """Enrich the imdb's roidb by adding some derived quantities that 15 | are useful for training. This function precomputes the maximum 16 | overlap, taken over ground-truth boxes, between each ROI and 17 | each ground-truth box. The class with maximum overlap is also 18 | recorded. 19 | """ 20 | 21 | roidb = imdb.roidb 22 | #if not (imdb.name.startswith('coco')): 23 | sizes = [PIL.Image.open(imdb.image_path_at(i)).size 24 | for i in range(imdb.num_images)] 25 | 26 | for i in range(len(imdb.image_index)): 27 | roidb[i]['img_id'] = imdb.image_id_at(i) 28 | roidb[i]['image'] = imdb.image_path_at(i) 29 | #if not (imdb.name.startswith('coco')): 30 | roidb[i]['width'] = sizes[i][0] 31 | roidb[i]['height'] = sizes[i][1] 32 | # need gt_overlaps as a dense array for argmax 33 | gt_overlaps = roidb[i]['gt_overlaps'].toarray() 34 | # max overlap with gt over classes (columns) 35 | max_overlaps = gt_overlaps.max(axis=1) 36 | # gt class that had the max overlap 37 | max_classes = gt_overlaps.argmax(axis=1) 38 | roidb[i]['max_classes'] = max_classes 39 | roidb[i]['max_overlaps'] = max_overlaps 40 | # sanity checks 41 | # max overlap of 0 => class should be zero (background) 42 | zero_inds = np.where(max_overlaps == 0)[0] 43 | assert all(max_classes[zero_inds] == 0) 44 | # max overlap > 0 => class should not be zero (must be a fg class) 45 | nonzero_inds = np.where(max_overlaps > 0)[0] 46 | assert all(max_classes[nonzero_inds] != 0) 47 | 48 | 49 | def rank_roidb_ratio(roidb): 50 | # rank roidb based on the ratio between width and height. 51 | ratio_large = 2 # largest ratio to preserve. 52 | ratio_small = 0.5 # smallest ratio to preserve. 53 | 54 | ratio_list = [] 55 | for i in range(len(roidb)): 56 | width = roidb[i]['width'] 57 | height = roidb[i]['height'] 58 | ratio = width / float(height) 59 | 60 | if ratio > ratio_large: 61 | roidb[i]['need_crop'] = 1 62 | ratio = ratio_large 63 | elif ratio < ratio_small: 64 | roidb[i]['need_crop'] = 1 65 | ratio = ratio_small 66 | else: 67 | roidb[i]['need_crop'] = 0 68 | 69 | ratio_list.append(ratio) 70 | 71 | ratio_list = np.array(ratio_list) 72 | ratio_index = np.argsort(ratio_list) 73 | return ratio_list[ratio_index], ratio_index 74 | 75 | def filter_roidb(roidb): 76 | # filter the image without bounding box. 77 | print('before filtering, there are %d images...' % (len(roidb))) 78 | i = 0 79 | while i < len(roidb): 80 | if len(roidb[i]['boxes']) == 0: 81 | del roidb[i] 82 | i -= 1 83 | i += 1 84 | 85 | print('after filtering, there are %d images...' % (len(roidb))) 86 | return roidb 87 | 88 | def combined_roidb(imdb_names, training=True): 89 | """ 90 | Combine multiple roidbs 91 | """ 92 | 93 | def get_training_roidb(imdb): 94 | """Returns a roidb (Region of Interest database) for use in training.""" 95 | # if cfg.TRAIN.USE_FLIPPED: 96 | # print('Appending horizontally-flipped training examples...') 97 | # imdb.append_flipped_images() 98 | # print('done') 99 | 100 | print('Preparing training data...') 101 | 102 | prepare_roidb(imdb) 103 | #ratio_index = rank_roidb_ratio(imdb) 104 | print('done') 105 | 106 | return imdb.roidb 107 | 108 | def get_roidb(imdb_name): 109 | imdb = get_imdb(imdb_name) 110 | print('Loaded dataset `{:s}` for training'.format(imdb.name)) 111 | imdb.set_proposal_method(cfg.TRAIN.PROPOSAL_METHOD) 112 | print('Set proposal method: {:s}'.format(cfg.TRAIN.PROPOSAL_METHOD)) 113 | roidb = get_training_roidb(imdb) 114 | return roidb 115 | 116 | #print(imdb_names.split('+')) 117 | roidbs = [get_roidb(s) for s in imdb_names.split('+')] 118 | roidb = roidbs[0] 119 | 120 | if len(roidbs) > 1: 121 | for r in roidbs[1:]: 122 | roidb.extend(r) 123 | tmp = get_imdb(imdb_names.split('+')[1]) 124 | imdb = datasets.imdb.imdb(imdb_names, tmp.classes) 125 | else: 126 | imdb = get_imdb(imdb_names) 127 | 128 | if training: 129 | roidb = filter_roidb(roidb) 130 | 131 | ratio_list, ratio_index = rank_roidb_ratio(roidb) 132 | 133 | return imdb, roidb, ratio_list, ratio_index 134 | -------------------------------------------------------------------------------- /lib/setup.py: -------------------------------------------------------------------------------- 1 | from __future__ import print_function 2 | # -------------------------------------------------------- 3 | # Fast R-CNN 4 | # Copyright (c) 2015 Microsoft 5 | # Licensed under The MIT License [see LICENSE for details] 6 | # Written by Ross Girshick 7 | # -------------------------------------------------------- 8 | 9 | import os 10 | from os.path import join as pjoin 11 | import numpy as np 12 | from distutils.core import setup 13 | from distutils.extension import Extension 14 | from Cython.Distutils import build_ext 15 | 16 | 17 | def find_in_path(name, path): 18 | "Find a file in a search path" 19 | # adapted fom http://code.activestate.com/recipes/52224-find-a-file-given-a-search-path/ 20 | for dir in path.split(os.pathsep): 21 | binpath = pjoin(dir, name) 22 | if os.path.exists(binpath): 23 | return os.path.abspath(binpath) 24 | return None 25 | 26 | 27 | # def locate_cuda(): 28 | # """Locate the CUDA environment on the system 29 | # 30 | # Returns a dict with keys 'home', 'nvcc', 'include', and 'lib64' 31 | # and values giving the absolute path to each directory. 32 | # 33 | # Starts by looking for the CUDAHOME env variable. If not found, everything 34 | # is based on finding 'nvcc' in the PATH. 35 | # """ 36 | # 37 | # # first check if the CUDAHOME env variable is in use 38 | # if 'CUDAHOME' in os.environ: 39 | # home = os.environ['CUDAHOME'] 40 | # nvcc = pjoin(home, 'bin', 'nvcc') 41 | # else: 42 | # # otherwise, search the PATH for NVCC 43 | # default_path = pjoin(os.sep, 'usr', 'local', 'cuda', 'bin') 44 | # nvcc = find_in_path('nvcc', os.environ['PATH'] + os.pathsep + default_path) 45 | # if nvcc is None: 46 | # raise EnvironmentError('The nvcc binary could not be ' 47 | # 'located in your $PATH. Either add it to your path, or set $CUDAHOME') 48 | # home = os.path.dirname(os.path.dirname(nvcc)) 49 | # 50 | # cudaconfig = {'home': home, 'nvcc': nvcc, 51 | # 'include': pjoin(home, 'include'), 52 | # 'lib64': pjoin(home, 'lib64')} 53 | # for k, v in cudaconfig.iteritems(): 54 | # if not os.path.exists(v): 55 | # raise EnvironmentError('The CUDA %s path could not be located in %s' % (k, v)) 56 | # 57 | # return cudaconfig 58 | 59 | 60 | # CUDA = locate_cuda() 61 | 62 | # Obtain the numpy include directory. This logic works across numpy versions. 63 | try: 64 | numpy_include = np.get_include() 65 | except AttributeError: 66 | numpy_include = np.get_numpy_include() 67 | 68 | 69 | def customize_compiler_for_nvcc(self): 70 | """inject deep into distutils to customize how the dispatch 71 | to gcc/nvcc works. 72 | 73 | If you subclass UnixCCompiler, it's not trivial to get your subclass 74 | injected in, and still have the right customizations (i.e. 75 | distutils.sysconfig.customize_compiler) run on it. So instead of going 76 | the OO route, I have this. Note, it's kindof like a wierd functional 77 | subclassing going on.""" 78 | 79 | # tell the compiler it can processes .cu 80 | self.src_extensions.append('.cu') 81 | 82 | # save references to the default compiler_so and _comple methods 83 | default_compiler_so = self.compiler_so 84 | super = self._compile 85 | 86 | # now redefine the _compile method. This gets executed for each 87 | # object but distutils doesn't have the ability to change compilers 88 | # based on source extension: we add it. 89 | def _compile(obj, src, ext, cc_args, extra_postargs, pp_opts): 90 | print(extra_postargs) 91 | if os.path.splitext(src)[1] == '.cu': 92 | # use the cuda for .cu files 93 | self.set_executable('compiler_so', CUDA['nvcc']) 94 | # use only a subset of the extra_postargs, which are 1-1 translated 95 | # from the extra_compile_args in the Extension class 96 | postargs = extra_postargs['nvcc'] 97 | else: 98 | postargs = extra_postargs['gcc'] 99 | 100 | super(obj, src, ext, cc_args, postargs, pp_opts) 101 | # reset the default compiler_so, which we might have changed for cuda 102 | self.compiler_so = default_compiler_so 103 | 104 | # inject our redefined _compile method into the class 105 | self._compile = _compile 106 | 107 | 108 | # run the customize_compiler 109 | class custom_build_ext(build_ext): 110 | def build_extensions(self): 111 | customize_compiler_for_nvcc(self.compiler) 112 | build_ext.build_extensions(self) 113 | 114 | 115 | ext_modules = [ 116 | Extension( 117 | "model.utils.cython_bbox", 118 | ["model/utils/bbox.pyx"], 119 | extra_compile_args={'gcc': ["-Wno-cpp", "-Wno-unused-function"]}, 120 | include_dirs=[numpy_include] 121 | ), 122 | Extension( 123 | 'pycocotools._mask', 124 | sources=['pycocotools/maskApi.c', 'pycocotools/_mask.pyx'], 125 | include_dirs=[numpy_include, 'pycocotools'], 126 | extra_compile_args={ 127 | 'gcc': ['-Wno-cpp', '-Wno-unused-function', '-std=c99']}, 128 | ), 129 | ] 130 | 131 | setup( 132 | name='faster_rcnn', 133 | ext_modules=ext_modules, 134 | # inject our custom trigger 135 | cmdclass={'build_ext': custom_build_ext}, 136 | ) 137 | -------------------------------------------------------------------------------- /requirements.txt: -------------------------------------------------------------------------------- 1 | cython 2 | cffi 3 | opencv-python 4 | scipy 5 | msgpack 6 | easydict 7 | matplotlib 8 | pyyaml 9 | tensorboardX 10 | --------------------------------------------------------------------------------