├── .gitignore ├── README.md ├── _init_paths.py ├── cfgs ├── da_res50.yml ├── res101.yml ├── res50.yml └── vgg16.yml ├── demo.py ├── lib ├── datasets │ ├── LISA_eval.py │ ├── LISA_voc.py │ ├── VOCdevkit-matlab-wrapper │ │ ├── get_voc_opts.m │ │ ├── voc_eval.m │ │ └── xVOCap.m │ ├── __init__.py │ ├── coco.py │ ├── cross_domain_eval.py │ ├── cross_domain_voc.py │ ├── datasets_info.py │ ├── deeplesion_eval.py │ ├── deeplesion_voc.py │ ├── dota_eval.py │ ├── dota_voc.py │ ├── ds_utils.py │ ├── factory.py │ ├── imagenet.py │ ├── imdb.py │ ├── kitchen_eval.py │ ├── kitchen_voc.py │ ├── kitti_eval.py │ ├── kitti_voc.py │ ├── pascal_voc.py │ ├── pascal_voc_rbg.py │ ├── tools │ │ └── mcg_munge.py │ ├── vg.py │ ├── vg_eval.py │ ├── voc_eval.py │ ├── widerface_eval.py │ └── widerface_voc.py ├── make.sh ├── model │ ├── __init__.py │ ├── faster_rcnn │ │ ├── DAResNet.py │ │ ├── SEResNet.py │ │ ├── __init__.py │ │ ├── domain_attention_module.py │ │ ├── faster_rcnn.py │ │ ├── faster_rcnn_uni.py │ │ ├── resnet.py │ │ ├── resnet_uni.py │ │ ├── se_module.py │ │ └── se_module_vector.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.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 │ │ ├── proposal_target_layer_cascade_caltech.py │ │ ├── rpn.py │ │ └── rpn_universal.py │ └── utils │ │ ├── .gitignore │ │ ├── __init__.py │ │ ├── bbox.pyx │ │ ├── blob.py │ │ ├── config.py │ │ ├── logger.py │ │ └── net_utils.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 └── training_info.py ├── requirements.txt ├── scripts ├── test_universal.sh └── train_universal.sh ├── test_universal.py └── universal_model.py /.gitignore: -------------------------------------------------------------------------------- 1 | # Ignore data folder 2 | data/ 3 | 4 | # Ignore models folder 5 | models/ 6 | 7 | # Ignore .pyc .so .o .pkl .pth .pth.tar files 8 | *.pyc 9 | *.so 10 | *.o 11 | *.pkl 12 | *.pth 13 | *.pth.tar 14 | 15 | # Ignore cache files 16 | __pycache__/ 17 | -------------------------------------------------------------------------------- /README.md: -------------------------------------------------------------------------------- 1 | # Towards Universal Object Detection by Domain Attention. 2 | 3 | by Xudong Wang, Zhaowei Cai, Dashan Gao and Nuno Vasconcelos in UC San Diego and 12 Sigma Technologies. 4 | 5 | This project is based on Pytorch reproduced Faster R-CNN by [jwyang](https://github.com/jwyang/faster-rcnn.pytorch) 6 | 7 | ### Project Pages 8 | http://www.svcl.ucsd.edu/projects/universal-detection/ 9 | 10 | ### Introduction 11 | This is the benchmark introduced in CVPR 2019 paper: [Towards Universal Object Detection by Domain Attention](https://arxiv.org/pdf/1904.04402.pdf). The goal of this benchmark is to encourage designing universal object detection system, capble of solving various detection tasks. To train and evaluate universal/multi-domain object detection systems, we established a new universal object detection benchmark (UODB) of 11 datasets. Details of UODB can be obtained from project pages. You can also download these datasets in that project page. 12 | 13 | ### Datasets Preparation 14 | 15 | First of all, clone the code 16 | 17 | Then, create a folder: 18 | ``` 19 | cd towards-universal-object-detection && mkdir data 20 | ``` 21 | 22 | Then put all the donwloaded datasets from [UODB benchmark](http://www.svcl.ucsd.edu/projects/universal-detection/) inside data folder and unzip all of them. 23 | 24 | All VOC format datasets should be in the structure of: 25 | 26 | datasets 27 | --Annotations 28 | --0.xml 29 | --ImageSets 30 | --Main 31 | --JPEGImages 32 | --0.jpg or 0.png 33 | 34 | ### prerequisites 35 | 36 | * Python 3.6 37 | * Pytorch 0.4.0 38 | * CUDA 8.0 or higher 39 | 40 | ### Pretrained Model 41 | 42 | You can download pre-trained models in ImageNet from: 43 | 44 | * DA-50: [Dropbox](https://drive.google.com/file/d/1kddC55_eByFfMZqDTM9cLj0j1BiHBq9D/view?usp=sharing) 45 | * ResNet50: [Dropbox](https://drive.google.com/file/d/1_0wFe2soxLkyP5DCCpOJddp1k_xcowv-/view?usp=sharing) 46 | 47 | Download and unzip DAResNet50.zip, and put them into data/pretrained_model/ folder. 48 | 49 | ### Compilation 50 | 51 | 1. Create virtual envrionment and activate it: 52 | 53 | ``` 54 | conda create -n uodb python=3.6 -y 55 | conda activate uodb 56 | ``` 57 | 58 | 2. Install all the python dependencies using pip: 59 | ``` 60 | pip install -r requirements.txt --user 61 | ``` 62 | 63 | 3. Install pytorch0.4.0 with conda: 64 | ``` 65 | conda install pytorch=0.4.0 cuda80 cudatoolkit==8.0 -c pytorch 66 | ``` 67 | Please change cuda version accordingly. 68 | 69 | 4. Compile the cuda dependencies using following simple commands: 70 | 71 | ``` 72 | cd lib 73 | sh make.sh 74 | ``` 75 | As pointed out by [ruotianluo/pytorch-faster-rcnn](https://github.com/ruotianluo/pytorch-faster-rcnn), choose the right `-arch` in `make.sh` file, to compile the cuda code: 76 | 77 | | GPU model | Architecture | 78 | | ------------- | ------------- | 79 | | Tesla K80 (AWS p2.xlarge) | sm_37 | 80 | | TitanX (Maxwell/Pascal) | sm_52 | 81 | | GTX 960M | sm_50 | 82 | | GTX 1080 (Ti) | sm_61 | 83 | | Grid K520 (AWS g2.2xlarge) | sm_30 | 84 | | RTX 2080 (Ti) | sm_70 | 85 | 86 | More details about setting the architecture can be found [here](https://developer.nvidia.com/cuda-gpus) or [here](http://arnon.dk/matching-sm-architectures-arch-and-gencode-for-various-nvidia-cards/) 87 | 88 | ## Train 89 | 90 | Before training, set the right directory to save and load the trained models. Change the arguments "save_dir" and "load_dir" in universal_model.py to adapt to your environment. 91 | 92 | To train a model with 11 adapters, simply run: 93 | ``` 94 | bash scripts/train_universal.sh 95 | ``` 96 | Above, BATCH_SIZE and WORKER_NUMBER can be set adaptively according to your GPU memory size. Specify the specific GPU device ID(GPU_ID), network(net), data directory(DATA_DIR), number of adapters(num_adapters), model session(SESSION), checkepoch(EPOCH), checkpoint iterations(CHECKPOINT), list of datasets to train(datasets_list), using less domain attention blocks(less_blocks) and etc. before running train_universal.sh file. 97 | 98 | ## Test 99 | 100 | If you want to evaluate the detection performance of each datasets, download pre-trained model and put it in models/da_res50/universal/, then simply run: 101 | ``` 102 | bash scripts/test_universal.sh 103 | ``` 104 | Specify the specific GPU device ID(GPU_ID), network(net), data directory(DATA_DIR), number of adapters(num_adapters), model session(SESSION), checkepoch(EPOCH), checkpoint iterations(CHECKPOINT), datasets to test(datasets) and etc. before running test_universal.sh file. Only sigle GPU testing is supported. 105 | 106 | Pre-trained model will be named as faster_rcnn_universal_SESSION_EPOCH_CHECKPOINT.pth 107 | 108 | Results and models for 5 datasets universal model: 109 | 110 | | #Adapter | less_blocks | KITTI | VOC | Widerface | LISA | Kitchen | AVG | download | 111 | | ------------- | ------------- | ------------- | ------------- | ------------- | ------------- | ------------- | ------------- | ------------- | 112 | | 3 | True | 68.0 | 78.8 | 51.9 | 88.1 | 87.1 | 74.8 | [model](https://drive.google.com/file/d/1CbbdBHmyIoALOTBSjZzEKPDHK9QKdUus/view?usp=sharing) | 113 | | 5 | True | 67.9 | 79.2 | 52.2 | 87.5 | 88.5 | 75.1 | [model](https://drive.google.com/file/d/1x5Rd33yeUXicOEXH6TIqhfYBX8wqmDRI/view?usp=sharing) | 114 | | 7 | True | 68.2 | 79.9 | 52.1 | 89.7 | 88.0 | 75.6 | [model](https://drive.google.com/file/d/1gLQZGn6Vb-AzfzFmf7YCZLqE_Mfwzyq6/view?usp=sharing) | 115 | 116 | Results and models for 11 datasets universal model: 117 | 118 | | #Adapter | less_blocks | KITTI | VOC | Widerface | LISA | Kitchen | COCO | DOTA | DeepLesion | Comic | Clipart | Watercolor | AVG | download | 119 | | ------------- | ------------- | ------------- | ------------- | ------------- | ------------- | ------------- | ------------- | ------------- | ------------- | ------------- | ------------- | ------------- | ------------- | ------------- | 120 | | 6 | True | 67.6 | 82.7 | 51.8 | 87.9 | 88.7 | 46.8 | 57.0 | 54.8 | 52.6 | 54.6 | 58.2 | 63.9 | [model](https://drive.google.com/file/d/1uyQ-BX_p8T3HEaJrsvUTshmnIy80Et5M/view?usp=sharing) | 121 | | 8 | True | 68.0 | 82.4 | 51.3 | 87.6 | 90.0 | 47.0 | 56.3 | 53.4 | 53.4 | 55.8 | 60.6 | 64.2 | [model](https://drive.google.com/file/d/1WtthQFm_SEbMVcQnZnD8Xgm3n-msDw1B/view?usp=sharing) | 122 | 123 | ### Some popular problems 124 | 1. fatal error: cuda.h: No such file or directory: 125 | 126 | Export C_INCLUDE_PATH=/usr/local/cuda-8.0/include:${C_INCLUDE_PATH}, then run "sh make.sh" 127 | 128 | 2. RuntimeError: CUDNN_STATUS_EXECUTION_FAILED: 129 | 130 | Usually, this is caused by using different cudnn when building and running pytorch. You can check this simply by running: torch.backends.cudnn.version(). You can also test by checking if the output of "pytorch.version.cuda" and "nvcc --version" gives you the same cudnn version. If the above checks fail, you need to reinstall pytorch and make sure to use the same cudnn within the inference time. 131 | 132 | 3. THCudaCheck FAIL file=/opt/conda/conda-bld/pytorch_1524586445097/work/aten/src/THC/THCGeneral.cpp line=844 error=11 : invalid argument 133 | 134 | This error will appear for RTX2080 GPU cards with cuda8.x or cuda9.x, you may need to install pytorch from source to solve it. Check [issue](https://github.com/pytorch/pytorch/issues/15797) and [issue](https://discuss.pytorch.org/t/thcudacheck-fail-file-pytorch-aten-src-thc-thcgeneral-cpp/31788/13) for details. This error can be ignored within inference time. 135 | 136 | If you meet any problems, please feel free to contact me by: frank.xudongwang@gmail.com 137 | 138 | ### Citation 139 | 140 | If you use our code/model/data, please cite our paper: 141 | 142 | @inproceedings{wang2019towards, 143 | title={Towards universal object detection by domain attention}, 144 | author={Wang, Xudong and Cai, Zhaowei and Gao, Dashan and Vasconcelos, Nuno}, 145 | booktitle={Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition}, 146 | pages={7289--7298}, 147 | year={2019} 148 | } 149 | -------------------------------------------------------------------------------- /_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/da_res50.yml: -------------------------------------------------------------------------------- 1 | EXP_DIR: da_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: da_res50_faster_rcnn 15 | TEST: 16 | HAS_RPN: True 17 | POOLING_MODE: crop # original crop -------------------------------------------------------------------------------- /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 | WEIGHT_DECAY: 0.0001 12 | DOUBLE_BIAS: False 13 | LEARNING_RATE: 0.001 14 | TEST: 15 | HAS_RPN: True 16 | POOLING_SIZE: 7 17 | POOLING_MODE: align 18 | CROP_RESIZE_WITH_MAX_POOL: False 19 | -------------------------------------------------------------------------------- /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 # original 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 #ORIGINAL 256 7 | PROPOSAL_METHOD: gt 8 | BG_THRESH_LO: 0.0 9 | BATCH_SIZE: 256 #ORIGINAL 256 10 | LEARNING_RATE: 0.01 11 | #base_size:W 12 | 13 | TEST: 14 | HAS_RPN: True 15 | POOLING_MODE: align 16 | CROP_RESIZE_WITH_MAX_POOL: False -------------------------------------------------------------------------------- /lib/datasets/LISA_eval.py: -------------------------------------------------------------------------------- 1 | # -------------------------------------------------------- 2 | # Fast/er R-CNN 3 | # Licensed under The MIT License [see LICENSE for details] 4 | # Written by Bharath Hariharan 5 | # -------------------------------------------------------- 6 | from __future__ import absolute_import 7 | from __future__ import division 8 | from __future__ import print_function 9 | 10 | import xml.etree.ElementTree as ET 11 | import os 12 | import pickle 13 | import numpy as np 14 | 15 | def parse_rec(filename): 16 | """ Parse a PASCAL VOC xml file """ 17 | tree = ET.parse(filename) 18 | objects = [] 19 | for obj in tree.findall('object'): 20 | obj_struct = {} 21 | obj_struct['name'] = obj.find('name').text 22 | # obj_struct['pose'] = obj.find('pose').text 23 | obj_struct['truncated'] = int(obj.find('truncated').text) 24 | obj_struct['difficult'] = int(obj.find('difficult').text) 25 | bbox = obj.find('bndbox') 26 | obj_struct['bbox'] = [int(bbox.find('xmin').text), 27 | int(bbox.find('ymin').text), 28 | int(bbox.find('xmax').text), 29 | int(bbox.find('ymax').text)] 30 | objects.append(obj_struct) 31 | 32 | return objects 33 | 34 | 35 | def voc_ap(rec, prec, use_07_metric=False): 36 | """ ap = voc_ap(rec, prec, [use_07_metric]) 37 | Compute VOC AP given precision and recall. 38 | If use_07_metric is true, uses the 39 | VOC 07 11 point method (default:False). 40 | """ 41 | if use_07_metric: 42 | # 11 point metric 43 | ap = 0. 44 | for t in np.arange(0., 1.1, 0.1): 45 | if np.sum(rec >= t) == 0: 46 | p = 0 47 | else: 48 | p = np.max(prec[rec >= t]) 49 | ap = ap + p / 11. 50 | else: 51 | # correct AP calculation 52 | # first append sentinel values at the end 53 | mrec = np.concatenate(([0.], rec, [1.])) 54 | mpre = np.concatenate(([0.], prec, [0.])) 55 | 56 | # compute the precision envelope 57 | for i in range(mpre.size - 1, 0, -1): 58 | mpre[i - 1] = np.maximum(mpre[i - 1], mpre[i]) 59 | 60 | # to calculate area under PR curve, look for points 61 | # where X axis (recall) changes value 62 | i = np.where(mrec[1:] != mrec[:-1])[0] 63 | 64 | # and sum (\Delta recall) * prec 65 | ap = np.sum((mrec[i + 1] - mrec[i]) * mpre[i + 1]) 66 | return ap 67 | 68 | 69 | def LISA_eval(detpath, 70 | annopath, 71 | imagesetfile, 72 | classname, 73 | cachedir, 74 | ovthresh=0.5, 75 | use_07_metric=False): 76 | """rec, prec, ap = voc_eval(detpath, 77 | annopath, 78 | imagesetfile, 79 | classname, 80 | [ovthresh], 81 | [use_07_metric]) 82 | 83 | Top level function that does the PASCAL VOC evaluation. 84 | 85 | detpath: Path to detections 86 | detpath.format(classname) should produce the detection results file. 87 | annopath: Path to annotations 88 | annopath.format(imagename) should be the xml annotations file. 89 | imagesetfile: Text file containing the list of images, one image per line. 90 | classname: Category name (duh) 91 | cachedir: Directory for caching the annotations 92 | [ovthresh]: Overlap threshold (default = 0.5) 93 | [use_07_metric]: Whether to use VOC07's 11 point AP computation 94 | (default False) 95 | """ 96 | # assumes detections are in detpath.format(classname) 97 | # assumes annotations are in annopath.format(imagename) 98 | # assumes imagesetfile is a text file with each line an image name 99 | # cachedir caches the annotations in a pickle file 100 | 101 | # first load gt 102 | if not os.path.isdir(cachedir): 103 | os.mkdir(cachedir) 104 | cachefile = os.path.join(cachedir, '%s_annots.pkl' % imagesetfile.split('/')[-1].split('.txt')[0]) 105 | # read list of images 106 | with open(imagesetfile, 'r') as f: 107 | lines = f.readlines() 108 | imagenames = [x.strip() for x in lines] 109 | 110 | if not os.path.isfile(cachefile): 111 | # load annotations 112 | recs = {} 113 | for i, imagename in enumerate(imagenames): 114 | recs[imagename] = parse_rec(annopath.format(imagename)) 115 | # if i % 100 == 0: 116 | # print('Reading annotation for {:d}/{:d}'.format( 117 | # i + 1, len(imagenames))) 118 | # save 119 | print('Saving cached annotations to {:s}'.format(cachefile)) 120 | with open(cachefile, 'wb') as f: 121 | pickle.dump(recs, f) 122 | else: 123 | # load 124 | with open(cachefile, 'rb') as f: 125 | try: 126 | recs = pickle.load(f) 127 | except: 128 | recs = pickle.load(f, encoding='bytes') 129 | 130 | # extract gt objects for this class 131 | class_recs = {} 132 | npos = 0 133 | for imagename in imagenames: 134 | R = [obj for obj in recs[imagename] if obj['name'] == classname] 135 | bbox = np.array([x['bbox'] for x in R]) 136 | difficult = np.array([x['difficult'] for x in R]).astype(np.bool) 137 | det = [False] * len(R) 138 | npos = npos + sum(~difficult) 139 | class_recs[imagename] = {'bbox': bbox, 140 | 'difficult': difficult, 141 | 'det': det} 142 | 143 | # read dets 144 | detfile = detpath.format(classname) 145 | with open(detfile, 'r') as f: 146 | lines = f.readlines() 147 | 148 | splitlines = [x.strip().split(' ') for x in lines] 149 | image_ids = [x[0] for x in splitlines] 150 | confidence = np.array([float(x[1]) for x in splitlines]) 151 | BB = np.array([[float(z) for z in x[2:]] for x in splitlines]) 152 | 153 | nd = len(image_ids) 154 | tp = np.zeros(nd) 155 | fp = np.zeros(nd) 156 | 157 | if BB.shape[0] > 0: 158 | # sort by confidence 159 | sorted_ind = np.argsort(-confidence) 160 | sorted_scores = np.sort(-confidence) 161 | BB = BB[sorted_ind, :] 162 | image_ids = [image_ids[x] for x in sorted_ind] 163 | 164 | # go down dets and mark TPs and FPs 165 | for d in range(nd): 166 | R = class_recs[image_ids[d]] 167 | bb = BB[d, :].astype(float) 168 | ovmax = -np.inf 169 | BBGT = R['bbox'].astype(float) 170 | 171 | if BBGT.size > 0: 172 | # compute overlaps 173 | # intersection 174 | ixmin = np.maximum(BBGT[:, 0], bb[0]) 175 | iymin = np.maximum(BBGT[:, 1], bb[1]) 176 | ixmax = np.minimum(BBGT[:, 2], bb[2]) 177 | iymax = np.minimum(BBGT[:, 3], bb[3]) 178 | iw = np.maximum(ixmax - ixmin + 1., 0.) 179 | ih = np.maximum(iymax - iymin + 1., 0.) 180 | inters = iw * ih 181 | 182 | # union 183 | uni = ((bb[2] - bb[0] + 1.) * (bb[3] - bb[1] + 1.) + 184 | (BBGT[:, 2] - BBGT[:, 0] + 1.) * 185 | (BBGT[:, 3] - BBGT[:, 1] + 1.) - inters) 186 | 187 | overlaps = inters / uni 188 | ovmax = np.max(overlaps) 189 | jmax = np.argmax(overlaps) 190 | 191 | if ovmax > ovthresh: 192 | if not R['difficult'][jmax]: 193 | if not R['det'][jmax]: 194 | tp[d] = 1. 195 | R['det'][jmax] = 1 196 | else: 197 | fp[d] = 1. 198 | else: 199 | fp[d] = 1. 200 | 201 | # compute precision recall 202 | fp = np.cumsum(fp) 203 | tp = np.cumsum(tp) 204 | rec = tp / float(npos) 205 | # avoid divide by zero in case the first detection matches a difficult 206 | # ground truth 207 | prec = tp / np.maximum(tp + fp, np.finfo(np.float64).eps) 208 | ap = voc_ap(rec, prec, use_07_metric) 209 | 210 | return rec, prec, ap 211 | -------------------------------------------------------------------------------- /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/cross_domain_eval.py: -------------------------------------------------------------------------------- 1 | # -------------------------------------------------------- 2 | # Fast/er R-CNN 3 | # Licensed under The MIT License [see LICENSE for details] 4 | # Written by Bharath Hariharan 5 | # -------------------------------------------------------- 6 | from __future__ import absolute_import 7 | from __future__ import division 8 | from __future__ import print_function 9 | 10 | import xml.etree.ElementTree as ET 11 | import os 12 | import pickle 13 | import numpy as np 14 | 15 | def parse_rec(filename): 16 | """ Parse a PASCAL VOC xml file """ 17 | tree = ET.parse(filename) 18 | objects = [] 19 | for obj in tree.findall('object'): 20 | obj_struct = {} 21 | obj_struct['name'] = obj.find('name').text 22 | obj_struct['pose'] = obj.find('pose').text 23 | obj_struct['truncated'] = int(obj.find('truncated').text) 24 | obj_struct['difficult'] = int(obj.find('difficult').text) 25 | bbox = obj.find('bndbox') 26 | obj_struct['bbox'] = [int(bbox.find('xmin').text), 27 | int(bbox.find('ymin').text), 28 | int(bbox.find('xmax').text), 29 | int(bbox.find('ymax').text)] 30 | objects.append(obj_struct) 31 | 32 | return objects 33 | 34 | 35 | def voc_ap(rec, prec, use_07_metric=False): 36 | """ ap = voc_ap(rec, prec, [use_07_metric]) 37 | Compute VOC AP given precision and recall. 38 | If use_07_metric is true, uses the 39 | VOC 07 11 point method (default:False). 40 | """ 41 | if use_07_metric: 42 | # 11 point metric 43 | ap = 0. 44 | for t in np.arange(0., 1.1, 0.1): 45 | if np.sum(rec >= t) == 0: 46 | p = 0 47 | else: 48 | p = np.max(prec[rec >= t]) 49 | ap = ap + p / 11. 50 | else: 51 | # correct AP calculation 52 | # first append sentinel values at the end 53 | mrec = np.concatenate(([0.], rec, [1.])) 54 | mpre = np.concatenate(([0.], prec, [0.])) 55 | 56 | # compute the precision envelope 57 | for i in range(mpre.size - 1, 0, -1): 58 | mpre[i - 1] = np.maximum(mpre[i - 1], mpre[i]) 59 | 60 | # to calculate area under PR curve, look for points 61 | # where X axis (recall) changes value 62 | i = np.where(mrec[1:] != mrec[:-1])[0] 63 | 64 | # and sum (\Delta recall) * prec 65 | ap = np.sum((mrec[i + 1] - mrec[i]) * mpre[i + 1]) 66 | return ap 67 | 68 | 69 | def voc_eval(detpath, 70 | annopath, 71 | imagesetfile, 72 | classname, 73 | cachedir, 74 | ovthresh=0.5, 75 | use_07_metric=False): 76 | """rec, prec, ap = voc_eval(detpath, 77 | annopath, 78 | imagesetfile, 79 | classname, 80 | [ovthresh], 81 | [use_07_metric]) 82 | 83 | Top level function that does the PASCAL VOC evaluation. 84 | 85 | detpath: Path to detections 86 | detpath.format(classname) should produce the detection results file. 87 | annopath: Path to annotations 88 | annopath.format(imagename) should be the xml annotations file. 89 | imagesetfile: Text file containing the list of images, one image per line. 90 | classname: Category name (duh) 91 | cachedir: Directory for caching the annotations 92 | [ovthresh]: Overlap threshold (default = 0.5) 93 | [use_07_metric]: Whether to use VOC07's 11 point AP computation 94 | (default False) 95 | """ 96 | # assumes detections are in detpath.format(classname) 97 | # assumes annotations are in annopath.format(imagename) 98 | # assumes imagesetfile is a text file with each line an image name 99 | # cachedir caches the annotations in a pickle file 100 | 101 | # first load gt 102 | if not os.path.isdir(cachedir): 103 | os.mkdir(cachedir) 104 | cachefile = os.path.join(cachedir, '%s_annots.pkl' % imagesetfile.split('/')[-1].split('.txt')[0]) 105 | # read list of images 106 | with open(imagesetfile, 'r') as f: 107 | lines = f.readlines() 108 | imagenames = [x.strip() for x in lines] 109 | 110 | if not os.path.isfile(cachefile): 111 | # load annotations 112 | recs = {} 113 | for i, imagename in enumerate(imagenames): 114 | recs[imagename] = parse_rec(annopath.format(imagename)) 115 | # if i % 100 == 0: 116 | # print('Reading annotation for {:d}/{:d}'.format( 117 | # i + 1, len(imagenames))) 118 | # save 119 | print('Saving cached annotations to {:s}'.format(cachefile)) 120 | with open(cachefile, 'wb') as f: 121 | pickle.dump(recs, f) 122 | else: 123 | # load 124 | with open(cachefile, 'rb') as f: 125 | try: 126 | recs = pickle.load(f) 127 | except: 128 | recs = pickle.load(f, encoding='bytes') 129 | 130 | # extract gt objects for this class 131 | class_recs = {} 132 | npos = 0 133 | for imagename in imagenames: 134 | R = [obj for obj in recs[imagename] if obj['name'] == classname] 135 | bbox = np.array([x['bbox'] for x in R]) 136 | difficult = np.array([x['difficult'] for x in R]).astype(np.bool) 137 | det = [False] * len(R) 138 | npos = npos + sum(~difficult) 139 | class_recs[imagename] = {'bbox': bbox, 140 | 'difficult': difficult, 141 | 'det': det} 142 | 143 | # read dets 144 | detfile = detpath.format(classname) 145 | with open(detfile, 'r') as f: 146 | lines = f.readlines() 147 | 148 | splitlines = [x.strip().split(' ') for x in lines] 149 | image_ids = [x[0] for x in splitlines] 150 | confidence = np.array([float(x[1]) for x in splitlines]) 151 | BB = np.array([[float(z) for z in x[2:]] for x in splitlines]) 152 | 153 | nd = len(image_ids) 154 | tp = np.zeros(nd) 155 | fp = np.zeros(nd) 156 | 157 | if BB.shape[0] > 0: 158 | # sort by confidence 159 | sorted_ind = np.argsort(-confidence) 160 | sorted_scores = np.sort(-confidence) 161 | BB = BB[sorted_ind, :] 162 | image_ids = [image_ids[x] for x in sorted_ind] 163 | 164 | # go down dets and mark TPs and FPs 165 | for d in range(nd): 166 | R = class_recs[image_ids[d]] 167 | bb = BB[d, :].astype(float) 168 | ovmax = -np.inf 169 | BBGT = R['bbox'].astype(float) 170 | 171 | if BBGT.size > 0: 172 | # compute overlaps 173 | # intersection 174 | ixmin = np.maximum(BBGT[:, 0], bb[0]) 175 | iymin = np.maximum(BBGT[:, 1], bb[1]) 176 | ixmax = np.minimum(BBGT[:, 2], bb[2]) 177 | iymax = np.minimum(BBGT[:, 3], bb[3]) 178 | iw = np.maximum(ixmax - ixmin + 1., 0.) 179 | ih = np.maximum(iymax - iymin + 1., 0.) 180 | inters = iw * ih 181 | 182 | # union 183 | uni = ((bb[2] - bb[0] + 1.) * (bb[3] - bb[1] + 1.) + 184 | (BBGT[:, 2] - BBGT[:, 0] + 1.) * 185 | (BBGT[:, 3] - BBGT[:, 1] + 1.) - inters) 186 | 187 | overlaps = inters / uni 188 | ovmax = np.max(overlaps) 189 | jmax = np.argmax(overlaps) 190 | 191 | if ovmax > ovthresh: 192 | if not R['difficult'][jmax]: 193 | if not R['det'][jmax]: 194 | tp[d] = 1. 195 | R['det'][jmax] = 1 196 | else: 197 | fp[d] = 1. 198 | else: 199 | fp[d] = 1. 200 | 201 | # compute precision recall 202 | fp = np.cumsum(fp) 203 | tp = np.cumsum(tp) 204 | rec = tp / float(npos) 205 | # avoid divide by zero in case the first detection matches a difficult 206 | # ground truth 207 | prec = tp / np.maximum(tp + fp, np.finfo(np.float64).eps) 208 | ap = voc_ap(rec, prec, use_07_metric) 209 | 210 | return rec, prec, ap -------------------------------------------------------------------------------- /lib/datasets/deeplesion_eval.py: -------------------------------------------------------------------------------- 1 | # -------------------------------------------------------- 2 | # Fast/er R-CNN 3 | # Licensed under The MIT License [see LICENSE for details] 4 | # Written by Bharath Hariharan 5 | # -------------------------------------------------------- 6 | from __future__ import absolute_import 7 | from __future__ import division 8 | from __future__ import print_function 9 | 10 | import xml.etree.ElementTree as ET 11 | import os 12 | import pickle 13 | import numpy as np 14 | 15 | def parse_rec(filename): 16 | """ Parse a PASCAL VOC xml file """ 17 | tree = ET.parse(filename) 18 | objects = [] 19 | for obj in tree.findall('object'): 20 | obj_struct = {} 21 | obj_struct['name'] = obj.find('name').text 22 | obj_struct['pose'] = obj.find('pose').text 23 | # obj_struct['truncated'] = int(obj.find('truncated').text) 24 | obj_struct['difficult'] = int(obj.find('difficult').text) 25 | bbox = obj.find('bndbox') 26 | obj_struct['bbox'] = [int(float(bbox.find('xmin').text)), 27 | int(float(bbox.find('ymin').text)), 28 | int(float(bbox.find('xmax').text)), 29 | int(float(bbox.find('ymax').text))] 30 | objects.append(obj_struct) 31 | 32 | return objects 33 | 34 | 35 | def voc_ap(rec, prec, use_07_metric=False): 36 | """ ap = voc_ap(rec, prec, [use_07_metric]) 37 | Compute VOC AP given precision and recall. 38 | If use_07_metric is true, uses the 39 | VOC 07 11 point method (default:False). 40 | """ 41 | if use_07_metric: 42 | # 11 point metric 43 | ap = 0. 44 | for t in np.arange(0., 1.1, 0.1): 45 | if np.sum(rec >= t) == 0: 46 | p = 0 47 | else: 48 | p = np.max(prec[rec >= t]) 49 | ap = ap + p / 11. 50 | else: 51 | # correct AP calculation 52 | # first append sentinel values at the end 53 | mrec = np.concatenate(([0.], rec, [1.])) 54 | mpre = np.concatenate(([0.], prec, [0.])) 55 | 56 | # compute the precision envelope 57 | for i in range(mpre.size - 1, 0, -1): 58 | mpre[i - 1] = np.maximum(mpre[i - 1], mpre[i]) 59 | 60 | # to calculate area under PR curve, look for points 61 | # where X axis (recall) changes value 62 | i = np.where(mrec[1:] != mrec[:-1])[0] 63 | 64 | # and sum (\Delta recall) * prec 65 | ap = np.sum((mrec[i + 1] - mrec[i]) * mpre[i + 1]) 66 | return ap 67 | 68 | 69 | def voc_eval(detpath, 70 | annopath, 71 | imagesetfile, 72 | classname, 73 | cachedir, 74 | ovthresh=0.5, 75 | use_07_metric=False): 76 | """rec, prec, ap = voc_eval(detpath, 77 | annopath, 78 | imagesetfile, 79 | classname, 80 | [ovthresh], 81 | [use_07_metric]) 82 | 83 | Top level function that does the PASCAL VOC evaluation. 84 | 85 | detpath: Path to detections 86 | detpath.format(classname) should produce the detection results file. 87 | annopath: Path to annotations 88 | annopath.format(imagename) should be the xml annotations file. 89 | imagesetfile: Text file containing the list of images, one image per line. 90 | classname: Category name (duh) 91 | cachedir: Directory for caching the annotations 92 | [ovthresh]: Overlap threshold (default = 0.5) 93 | [use_07_metric]: Whether to use VOC07's 11 point AP computation 94 | (default False) 95 | """ 96 | # assumes detections are in detpath.format(classname) 97 | # assumes annotations are in annopath.format(imagename) 98 | # assumes imagesetfile is a text file with each line an image name 99 | # cachedir caches the annotations in a pickle file 100 | 101 | # first load gt 102 | if not os.path.isdir(cachedir): 103 | os.mkdir(cachedir) 104 | cachefile = os.path.join(cachedir, '%s_annots.pkl' % imagesetfile.split('/')[-1].split('.txt')[0]) 105 | # read list of images 106 | with open(imagesetfile, 'r') as f: 107 | lines = f.readlines() 108 | imagenames = [x.strip() for x in lines] 109 | 110 | if not os.path.isfile(cachefile): 111 | # load annotations 112 | recs = {} 113 | for i, imagename in enumerate(imagenames): 114 | recs[imagename] = parse_rec(annopath.format(imagename)) 115 | # if i % 100 == 0: 116 | # print('Reading annotation for {:d}/{:d}'.format( 117 | # i + 1, len(imagenames))) 118 | # save 119 | print('Saving cached annotations to {:s}'.format(cachefile)) 120 | with open(cachefile, 'wb') as f: 121 | pickle.dump(recs, f) 122 | else: 123 | # load 124 | with open(cachefile, 'rb') as f: 125 | try: 126 | recs = pickle.load(f) 127 | except: 128 | recs = pickle.load(f, encoding='bytes') 129 | 130 | # extract gt objects for this class 131 | class_recs = {} 132 | npos = 0 133 | for imagename in imagenames: 134 | R = [obj for obj in recs[imagename] if obj['name'] == classname] 135 | bbox = np.array([x['bbox'] for x in R]) 136 | difficult = np.array([x['difficult'] for x in R]).astype(np.bool) 137 | det = [False] * len(R) 138 | npos = npos + sum(~difficult) 139 | class_recs[imagename] = {'bbox': bbox, 140 | 'difficult': difficult, 141 | 'det': det} 142 | 143 | # read dets 144 | detfile = detpath.format(classname) 145 | with open(detfile, 'r') as f: 146 | lines = f.readlines() 147 | 148 | splitlines = [x.strip().split(' ') for x in lines] 149 | image_ids = [x[0] for x in splitlines] 150 | confidence = np.array([float(x[1]) for x in splitlines]) 151 | BB = np.array([[float(z) for z in x[2:]] for x in splitlines]) 152 | 153 | nd = len(image_ids) 154 | tp = np.zeros(nd) 155 | fp = np.zeros(nd) 156 | 157 | if BB.shape[0] > 0: 158 | # sort by confidence 159 | sorted_ind = np.argsort(-confidence) 160 | sorted_scores = np.sort(-confidence) 161 | BB = BB[sorted_ind, :] 162 | image_ids = [image_ids[x] for x in sorted_ind] 163 | 164 | # go down dets and mark TPs and FPs 165 | for d in range(nd): 166 | R = class_recs[image_ids[d]] 167 | bb = BB[d, :].astype(float) 168 | ovmax = -np.inf 169 | BBGT = R['bbox'].astype(float) 170 | 171 | if BBGT.size > 0: 172 | # compute overlaps 173 | # intersection 174 | ixmin = np.maximum(BBGT[:, 0], bb[0]) 175 | iymin = np.maximum(BBGT[:, 1], bb[1]) 176 | ixmax = np.minimum(BBGT[:, 2], bb[2]) 177 | iymax = np.minimum(BBGT[:, 3], bb[3]) 178 | iw = np.maximum(ixmax - ixmin + 1., 0.) 179 | ih = np.maximum(iymax - iymin + 1., 0.) 180 | inters = iw * ih 181 | 182 | # union 183 | uni = ((bb[2] - bb[0] + 1.) * (bb[3] - bb[1] + 1.) + 184 | (BBGT[:, 2] - BBGT[:, 0] + 1.) * 185 | (BBGT[:, 3] - BBGT[:, 1] + 1.) - inters) 186 | 187 | overlaps = inters / uni 188 | ovmax = np.max(overlaps) 189 | jmax = np.argmax(overlaps) 190 | 191 | if ovmax > ovthresh: 192 | if not R['difficult'][jmax]: 193 | if not R['det'][jmax]: 194 | tp[d] = 1. 195 | R['det'][jmax] = 1 196 | else: 197 | fp[d] = 1. 198 | else: 199 | fp[d] = 1. 200 | 201 | # compute precision recall 202 | fp = np.cumsum(fp) 203 | tp = np.cumsum(tp) 204 | rec = tp / float(npos) 205 | # avoid divide by zero in case the first detection matches a difficult 206 | # ground truth 207 | prec = tp / np.maximum(tp + fp, np.finfo(np.float64).eps) 208 | ap = voc_ap(rec, prec, use_07_metric) 209 | 210 | return rec, prec, ap 211 | -------------------------------------------------------------------------------- /lib/datasets/dota_eval.py: -------------------------------------------------------------------------------- 1 | # -------------------------------------------------------- 2 | # Fast/er R-CNN 3 | # Licensed under The MIT License [see LICENSE for details] 4 | # Written by Bharath Hariharan 5 | # -------------------------------------------------------- 6 | from __future__ import absolute_import 7 | from __future__ import division 8 | from __future__ import print_function 9 | 10 | import xml.etree.ElementTree as ET 11 | import os 12 | import pickle 13 | import numpy as np 14 | 15 | def parse_rec(filename): 16 | """ Parse a PASCAL VOC xml file """ 17 | tree = ET.parse(filename) 18 | objects = [] 19 | for obj in tree.findall('object'): 20 | obj_struct = {} 21 | obj_struct['name'] = obj.find('name').text 22 | # obj_struct['pose'] = obj.find('pose').text 23 | # obj_struct['truncated'] = int(obj.find('truncated').text) 24 | obj_struct['difficult'] = int(obj.find('difficult').text) 25 | bbox = obj.find('bndbox') 26 | obj_struct['bbox'] = [int(float(bbox.find('xmin').text)), 27 | int(float(bbox.find('ymin').text)), 28 | int(float(bbox.find('xmax').text)), 29 | int(float(bbox.find('ymax').text))] 30 | objects.append(obj_struct) 31 | 32 | return objects 33 | 34 | 35 | def voc_ap(rec, prec, use_07_metric=False): 36 | """ ap = voc_ap(rec, prec, [use_07_metric]) 37 | Compute VOC AP given precision and recall. 38 | If use_07_metric is true, uses the 39 | VOC 07 11 point method (default:False). 40 | """ 41 | if use_07_metric: 42 | # 11 point metric 43 | ap = 0. 44 | for t in np.arange(0., 1.1, 0.1): 45 | if np.sum(rec >= t) == 0: 46 | p = 0 47 | else: 48 | p = np.max(prec[rec >= t]) 49 | ap = ap + p / 11. 50 | else: 51 | # correct AP calculation 52 | # first append sentinel values at the end 53 | mrec = np.concatenate(([0.], rec, [1.])) 54 | mpre = np.concatenate(([0.], prec, [0.])) 55 | 56 | # compute the precision envelope 57 | for i in range(mpre.size - 1, 0, -1): 58 | mpre[i - 1] = np.maximum(mpre[i - 1], mpre[i]) 59 | 60 | # to calculate area under PR curve, look for points 61 | # where X axis (recall) changes value 62 | i = np.where(mrec[1:] != mrec[:-1])[0] 63 | 64 | # and sum (\Delta recall) * prec 65 | ap = np.sum((mrec[i + 1] - mrec[i]) * mpre[i + 1]) 66 | return ap 67 | 68 | 69 | def voc_eval(detpath, 70 | annopath, 71 | imagesetfile, 72 | classname, 73 | cachedir, 74 | ovthresh=0.5, 75 | use_07_metric=False): 76 | """rec, prec, ap = voc_eval(detpath, 77 | annopath, 78 | imagesetfile, 79 | classname, 80 | [ovthresh], 81 | [use_07_metric]) 82 | 83 | Top level function that does the PASCAL VOC evaluation. 84 | 85 | detpath: Path to detections 86 | detpath.format(classname) should produce the detection results file. 87 | annopath: Path to annotations 88 | annopath.format(imagename) should be the xml annotations file. 89 | imagesetfile: Text file containing the list of images, one image per line. 90 | classname: Category name (duh) 91 | cachedir: Directory for caching the annotations 92 | [ovthresh]: Overlap threshold (default = 0.5) 93 | [use_07_metric]: Whether to use VOC07's 11 point AP computation 94 | (default False) 95 | """ 96 | # assumes detections are in detpath.format(classname) 97 | # assumes annotations are in annopath.format(imagename) 98 | # assumes imagesetfile is a text file with each line an image name 99 | # cachedir caches the annotations in a pickle file 100 | 101 | # first load gt 102 | if not os.path.isdir(cachedir): 103 | os.mkdir(cachedir) 104 | cachefile = os.path.join(cachedir, '%s_annots.pkl' % imagesetfile.split('/')[-1].split('.txt')[0]) 105 | # read list of images 106 | with open(imagesetfile, 'r') as f: 107 | lines = f.readlines() 108 | imagenames = [x.strip() for x in lines] 109 | 110 | # load annotations 111 | recs = {} 112 | for i, imagename in enumerate(imagenames): 113 | recs[imagename] = parse_rec(annopath.format(imagename)) 114 | #if i % 100 == 0: 115 | #print('Reading annotation for {:d}/{:d}'.format( 116 | #i + 1, len(imagenames))) 117 | # save 118 | print('Saving cached annotations to {:s}'.format(cachefile)) 119 | # with open(cachefile, 'wb') as f: 120 | # pickle.dump(recs, f) 121 | # else: 122 | # # load 123 | # with open(cachefile, 'rb') as f: 124 | # try: 125 | # recs = pickle.load(f) 126 | # except: 127 | # recs = pickle.load(f, encoding='bytes') 128 | 129 | # extract gt objects for this class 130 | class_recs = {} 131 | npos = 0 132 | for imagename in imagenames: 133 | R = [obj for obj in recs[imagename] if obj['name'] == classname] 134 | bbox = np.array([x['bbox'] for x in R]) 135 | difficult = np.array([int(x['difficult']) == 1 for x in R]) 136 | det = [False] * len(R) 137 | if len(difficult) != 0: 138 | npos = npos + sum(~difficult) 139 | class_recs[imagename] = {'bbox': bbox, 140 | 'difficult': difficult, 141 | 'det': det} 142 | 143 | # read dets 144 | detfile = detpath.format(classname) 145 | with open(detfile, 'r') as f: 146 | lines = f.readlines() 147 | 148 | splitlines = [x.strip().split(' ') for x in lines] 149 | image_ids = [x[0] for x in splitlines] 150 | confidence = np.array([float(x[1]) for x in splitlines]) 151 | BB = np.array([[float(z) for z in x[2:]] for x in splitlines]) 152 | 153 | nd = len(image_ids) 154 | tp = np.zeros(nd) 155 | fp = np.zeros(nd) 156 | 157 | if BB.shape[0] > 0: 158 | # sort by confidence 159 | sorted_ind = np.argsort(-confidence) 160 | sorted_scores = np.sort(-confidence) 161 | BB = BB[sorted_ind, :] 162 | image_ids = [image_ids[x] for x in sorted_ind] 163 | 164 | # go down dets and mark TPs and FPs 165 | for d in range(nd): 166 | R = class_recs[image_ids[d]] 167 | bb = BB[d, :].astype(float) 168 | ovmax = -np.inf 169 | BBGT = R['bbox'].astype(float) 170 | 171 | if BBGT.size > 0: 172 | # compute overlaps 173 | # intersection 174 | ixmin = np.maximum(BBGT[:, 0], bb[0]) 175 | iymin = np.maximum(BBGT[:, 1], bb[1]) 176 | ixmax = np.minimum(BBGT[:, 2], bb[2]) 177 | iymax = np.minimum(BBGT[:, 3], bb[3]) 178 | iw = np.maximum(ixmax - ixmin + 1., 0.) 179 | ih = np.maximum(iymax - iymin + 1., 0.) 180 | inters = iw * ih 181 | 182 | # union 183 | uni = ((bb[2] - bb[0] + 1.) * (bb[3] - bb[1] + 1.) + 184 | (BBGT[:, 2] - BBGT[:, 0] + 1.) * 185 | (BBGT[:, 3] - BBGT[:, 1] + 1.) - inters) 186 | 187 | overlaps = inters / uni 188 | ovmax = np.max(overlaps) 189 | jmax = np.argmax(overlaps) 190 | 191 | if ovmax > ovthresh: 192 | if not R['difficult'][jmax]: 193 | if not R['det'][jmax]: 194 | tp[d] = 1. 195 | R['det'][jmax] = 1 196 | else: 197 | fp[d] = 1. 198 | else: 199 | fp[d] = 1. 200 | 201 | # compute precision recall 202 | fp = np.cumsum(fp) 203 | tp = np.cumsum(tp) 204 | rec = tp / float(npos) 205 | # avoid divide by zero in case the first detection matches a difficult 206 | # ground truth 207 | prec = tp / np.maximum(tp + fp, np.finfo(np.float64).eps) 208 | ap = voc_ap(rec, prec, use_07_metric) 209 | 210 | return rec, prec, ap 211 | -------------------------------------------------------------------------------- /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.kitti_voc import kitti_voc 16 | from datasets.widerface_voc import widerface_voc 17 | from datasets.coco import coco 18 | from datasets.imagenet import imagenet 19 | from datasets.cross_domain_voc import cross_domain 20 | from datasets.dota_voc import dota_voc 21 | from datasets.kitchen_voc import kitchen_voc 22 | from datasets.vg import vg 23 | from datasets.deeplesion_voc import deeplesion_voc 24 | from datasets.LISA_voc import LISA_voc 25 | from model.utils.config import cfg 26 | import os 27 | 28 | import numpy as np 29 | 30 | # Set up voc__ 31 | for year in ['2007', '2012']: 32 | for split in ['train', 'val', 'trainval', 'test']: 33 | name = 'voc_{}_{}'.format(year, split) 34 | __sets[name] = (lambda split=split, year=year: pascal_voc(split, year)) 35 | 36 | for split in ['train', 'val', 'trainval', 'test']: 37 | name = 'LISA_{}'.format(split) 38 | year='2007' 39 | __sets[name] = (lambda split=split, year=year: LISA_voc(split, year)) 40 | 41 | for split in ['train', 'val', 'trainval', 'test']: 42 | name = 'deeplesion_{}'.format(split) 43 | year='2007' 44 | __sets[name] = (lambda split=split, year=year: deeplesion_voc(split, year)) 45 | 46 | for split in ['train', 'val', 'trainval', 'test']: 47 | name = 'kitchen_{}'.format(split) 48 | year='2007' 49 | __sets[name] = (lambda split=split, year=year: kitchen_voc(split, year)) 50 | 51 | for split in ['train', 'val', 'trainval', 'test']: 52 | name = 'dota_{}'.format(split) 53 | year='2007' 54 | __sets[name] = (lambda split=split, year=year: dota_voc(split, year)) 55 | 56 | for split in ['train', 'val', 'trainval', 'test']: 57 | year = '2007' 58 | name = 'watercolor_{}'.format(split) 59 | __sets[name] = (lambda split=split, year=year, name=name: cross_domain(split, year, datasets='watercolor')) 60 | 61 | for split in ['train', 'val', 'trainval', 'test']: 62 | year = '2007' 63 | name = 'comic_{}'.format(split) 64 | __sets[name] = (lambda split=split, year=year, name=name: cross_domain(split, year, datasets='comic')) 65 | 66 | for split in ['train', 'val', 'trainval', 'test']: 67 | year = '2007' 68 | name = 'clipart_{}'.format(split) 69 | __sets[name] = (lambda split=split, year=year, name=name: cross_domain(split, year, datasets='clipart')) 70 | 71 | # for year in ['2007', '2012']: 72 | for split in ['train', 'val', 'trainval', 'test']: 73 | name = 'kittivoc_{}'.format(split) 74 | year='2007' 75 | __sets[name] = (lambda split=split, year=year: kitti_voc(split, year)) 76 | 77 | for split in ['train', 'val', 'trainval', 'test']: 78 | name = 'widerface_{}'.format(split) 79 | year = '2007' 80 | __sets[name] = (lambda split=split, year=year: widerface_voc(split, year)) 81 | 82 | # Set up coco_2014_ 83 | for year in ['2014']: 84 | for split in ['train', 'val', 'minival', 'valminusminival', 'trainval']: 85 | name = 'coco_{}_{}'.format(year, split) 86 | __sets[name] = (lambda split=split, year=year: coco(split, year)) 87 | 88 | # Set up coco_2014_cap_ 89 | for year in ['2014']: 90 | for split in ['train', 'val', 'capval', 'valminuscapval', 'trainval']: 91 | name = 'coco_{}_{}'.format(year, split) 92 | __sets[name] = (lambda split=split, year=year: coco(split, year)) 93 | 94 | # Set up coco_2015_ 95 | for year in ['2015']: 96 | for split in ['test', 'test-dev']: 97 | name = 'coco_{}_{}'.format(year, split) 98 | __sets[name] = (lambda split=split, year=year: coco(split, year)) 99 | 100 | # set up image net. 101 | for split in ['train', 'val', 'val1', 'val2', 'test']: 102 | name = 'imagenet_{}'.format(split) 103 | devkit_path = 'data/imagenet/ILSVRC/devkit' 104 | data_path = 'data/imagenet/ILSVRC' 105 | __sets[name] = (lambda split=split, devkit_path=devkit_path, data_path=data_path: imagenet(split,devkit_path,data_path)) 106 | 107 | def get_imdb(name): 108 | """Get an imdb (image database) by name.""" 109 | if name not in __sets: 110 | raise KeyError('Unknown dataset: {}'.format(name)) 111 | if cfg.dataset == 'cross_domain': 112 | return __sets[name](name=name) 113 | return __sets[name]() 114 | 115 | def list_imdbs(): 116 | """List all registered imdbs.""" 117 | return list(__sets.keys()) -------------------------------------------------------------------------------- /lib/datasets/kitchen_eval.py: -------------------------------------------------------------------------------- 1 | # -------------------------------------------------------- 2 | # Fast/er R-CNN 3 | # Licensed under The MIT License [see LICENSE for details] 4 | # Written by Bharath Hariharan 5 | # -------------------------------------------------------- 6 | from __future__ import absolute_import 7 | from __future__ import division 8 | from __future__ import print_function 9 | 10 | import xml.etree.ElementTree as ET 11 | import os 12 | import pickle 13 | import numpy as np 14 | 15 | def parse_rec(filename): 16 | """ Parse a PASCAL VOC xml file """ 17 | tree = ET.parse(filename) 18 | objects = [] 19 | for obj in tree.findall('object'): 20 | obj_struct = {} 21 | obj_struct['name'] = obj.find('name').text 22 | obj_struct['pose'] = obj.find('pose').text 23 | obj_struct['truncated'] = int(obj.find('truncated').text) 24 | obj_struct['difficult'] = int(obj.find('difficult').text) 25 | bbox = obj.find('bndbox') 26 | obj_struct['bbox'] = [int(bbox.find('xmin').text), 27 | int(bbox.find('ymin').text), 28 | int(bbox.find('xmax').text), 29 | int(bbox.find('ymax').text)] 30 | objects.append(obj_struct) 31 | 32 | return objects 33 | 34 | 35 | def voc_ap(rec, prec, use_07_metric=False): 36 | """ ap = voc_ap(rec, prec, [use_07_metric]) 37 | Compute VOC AP given precision and recall. 38 | If use_07_metric is true, uses the 39 | VOC 07 11 point method (default:False). 40 | """ 41 | if use_07_metric: 42 | # 11 point metric 43 | ap = 0. 44 | for t in np.arange(0., 1.1, 0.1): 45 | if np.sum(rec >= t) == 0: 46 | p = 0 47 | else: 48 | p = np.max(prec[rec >= t]) 49 | ap = ap + p / 11. 50 | else: 51 | # correct AP calculation 52 | # first append sentinel values at the end 53 | mrec = np.concatenate(([0.], rec, [1.])) 54 | mpre = np.concatenate(([0.], prec, [0.])) 55 | 56 | # compute the precision envelope 57 | for i in range(mpre.size - 1, 0, -1): 58 | mpre[i - 1] = np.maximum(mpre[i - 1], mpre[i]) 59 | 60 | # to calculate area under PR curve, look for points 61 | # where X axis (recall) changes value 62 | i = np.where(mrec[1:] != mrec[:-1])[0] 63 | 64 | # and sum (\Delta recall) * prec 65 | ap = np.sum((mrec[i + 1] - mrec[i]) * mpre[i + 1]) 66 | return ap 67 | 68 | 69 | def kitchen_eval(detpath, 70 | annopath, 71 | imagesetfile, 72 | classname, 73 | cachedir, 74 | ovthresh=0.5, 75 | use_07_metric=False): 76 | """rec, prec, ap = voc_eval(detpath, 77 | annopath, 78 | imagesetfile, 79 | classname, 80 | [ovthresh], 81 | [use_07_metric]) 82 | 83 | Top level function that does the PASCAL VOC evaluation. 84 | 85 | detpath: Path to detections 86 | detpath.format(classname) should produce the detection results file. 87 | annopath: Path to annotations 88 | annopath.format(imagename) should be the xml annotations file. 89 | imagesetfile: Text file containing the list of images, one image per line. 90 | classname: Category name (duh) 91 | cachedir: Directory for caching the annotations 92 | [ovthresh]: Overlap threshold (default = 0.5) 93 | [use_07_metric]: Whether to use VOC07's 11 point AP computation 94 | (default False) 95 | """ 96 | # assumes detections are in detpath.format(classname) 97 | # assumes annotations are in annopath.format(imagename) 98 | # assumes imagesetfile is a text file with each line an image name 99 | # cachedir caches the annotations in a pickle file 100 | 101 | # first load gt 102 | if not os.path.isdir(cachedir): 103 | os.mkdir(cachedir) 104 | cachefile = os.path.join(cachedir, '%s_annots.pkl' % imagesetfile.split('/')[-1].split('.txt')[0]) 105 | # read list of images 106 | with open(imagesetfile, 'r') as f: 107 | lines = f.readlines() 108 | imagenames = [x.strip() for x in lines] 109 | 110 | if not os.path.isfile(cachefile): 111 | # load annotations 112 | recs = {} 113 | for i, imagename in enumerate(imagenames): 114 | recs[imagename] = parse_rec(annopath.format(imagename)) 115 | # if i % 100 == 0: 116 | # print('Reading annotation for {:d}/{:d}'.format( 117 | # i + 1, len(imagenames))) 118 | # save 119 | print('Saving cached annotations to {:s}'.format(cachefile)) 120 | with open(cachefile, 'wb') as f: 121 | pickle.dump(recs, f) 122 | else: 123 | # load 124 | with open(cachefile, 'rb') as f: 125 | try: 126 | recs = pickle.load(f) 127 | except: 128 | recs = pickle.load(f, encoding='bytes') 129 | 130 | # extract gt objects for this class 131 | class_recs = {} 132 | npos = 0 133 | for imagename in imagenames: 134 | R = [obj for obj in recs[imagename] if obj['name'] == classname] 135 | bbox = np.array([x['bbox'] for x in R]) 136 | difficult = np.array([x['difficult'] for x in R]).astype(np.bool) 137 | det = [False] * len(R) 138 | npos = npos + sum(~difficult) 139 | class_recs[imagename] = {'bbox': bbox, 140 | 'difficult': difficult, 141 | 'det': det} 142 | 143 | # read dets 144 | detfile = detpath.format(classname) 145 | with open(detfile, 'r') as f: 146 | lines = f.readlines() 147 | 148 | splitlines = [x.strip().split(' ') for x in lines] 149 | image_ids = [x[0] for x in splitlines] 150 | confidence = np.array([float(x[1]) for x in splitlines]) 151 | BB = np.array([[float(z) for z in x[2:]] for x in splitlines]) 152 | 153 | nd = len(image_ids) 154 | tp = np.zeros(nd) 155 | fp = np.zeros(nd) 156 | 157 | if BB.shape[0] > 0: 158 | # sort by confidence 159 | sorted_ind = np.argsort(-confidence) 160 | sorted_scores = np.sort(-confidence) 161 | BB = BB[sorted_ind, :] 162 | image_ids = [image_ids[x] for x in sorted_ind] 163 | 164 | # go down dets and mark TPs and FPs 165 | for d in range(nd): 166 | R = class_recs[image_ids[d]] 167 | bb = BB[d, :].astype(float) 168 | ovmax = -np.inf 169 | BBGT = R['bbox'].astype(float) 170 | 171 | if BBGT.size > 0: 172 | # compute overlaps 173 | # intersection 174 | ixmin = np.maximum(BBGT[:, 0], bb[0]) 175 | iymin = np.maximum(BBGT[:, 1], bb[1]) 176 | ixmax = np.minimum(BBGT[:, 2], bb[2]) 177 | iymax = np.minimum(BBGT[:, 3], bb[3]) 178 | iw = np.maximum(ixmax - ixmin + 1., 0.) 179 | ih = np.maximum(iymax - iymin + 1., 0.) 180 | inters = iw * ih 181 | 182 | # union 183 | uni = ((bb[2] - bb[0] + 1.) * (bb[3] - bb[1] + 1.) + 184 | (BBGT[:, 2] - BBGT[:, 0] + 1.) * 185 | (BBGT[:, 3] - BBGT[:, 1] + 1.) - inters) 186 | 187 | overlaps = inters / uni 188 | ovmax = np.max(overlaps) 189 | jmax = np.argmax(overlaps) 190 | 191 | if ovmax > ovthresh: 192 | if not R['difficult'][jmax]: 193 | if not R['det'][jmax]: 194 | tp[d] = 1. 195 | R['det'][jmax] = 1 196 | else: 197 | fp[d] = 1. 198 | else: 199 | fp[d] = 1. 200 | 201 | # compute precision recall 202 | fp = np.cumsum(fp) 203 | tp = np.cumsum(tp) 204 | rec = tp / float(npos) 205 | # avoid divide by zero in case the first detection matches a difficult 206 | # ground truth 207 | prec = tp / np.maximum(tp + fp, np.finfo(np.float64).eps) 208 | ap = voc_ap(rec, prec, use_07_metric) 209 | 210 | return rec, prec, ap 211 | -------------------------------------------------------------------------------- /lib/datasets/kitti_eval.py: -------------------------------------------------------------------------------- 1 | # -------------------------------------------------------- 2 | # Fast/er R-CNN 3 | # Licensed under The MIT License [see LICENSE for details] 4 | # Written by Bharath Hariharan 5 | # -------------------------------------------------------- 6 | from __future__ import absolute_import 7 | from __future__ import division 8 | from __future__ import print_function 9 | 10 | import xml.etree.ElementTree as ET 11 | import os 12 | import pickle 13 | import numpy as np 14 | 15 | def parse_rec(filename): 16 | """ Parse a PASCAL VOC xml file """ 17 | tree = ET.parse(filename) 18 | objects = [] 19 | for obj in tree.findall('object'): 20 | obj_struct = {} 21 | obj_struct['name'] = obj.find('name').text 22 | # obj_struct['pose'] = obj.find('pose').text 23 | obj_struct['truncated'] = int(obj.find('truncated').text) 24 | obj_struct['difficult'] = int(obj.find('difficult').text) 25 | bbox = obj.find('bndbox') 26 | obj_struct['bbox'] = [int(bbox.find('xmin').text), 27 | int(bbox.find('ymin').text), 28 | int(bbox.find('xmax').text), 29 | int(bbox.find('ymax').text)] 30 | objects.append(obj_struct) 31 | 32 | return objects 33 | 34 | 35 | def voc_ap(rec, prec, use_07_metric=False): 36 | """ ap = voc_ap(rec, prec, [use_07_metric]) 37 | Compute VOC AP given precision and recall. 38 | If use_07_metric is true, uses the 39 | VOC 07 11 point method (default:False). 40 | """ 41 | if use_07_metric: 42 | # 11 point metric 43 | ap = 0. 44 | for t in np.arange(0., 1.1, 0.1): 45 | if np.sum(rec >= t) == 0: 46 | p = 0 47 | else: 48 | p = np.max(prec[rec >= t]) 49 | ap = ap + p / 11. 50 | else: 51 | # correct AP calculation 52 | # first append sentinel values at the end 53 | mrec = np.concatenate(([0.], rec, [1.])) 54 | mpre = np.concatenate(([0.], prec, [0.])) 55 | 56 | # compute the precision envelope 57 | for i in range(mpre.size - 1, 0, -1): 58 | mpre[i - 1] = np.maximum(mpre[i - 1], mpre[i]) 59 | 60 | # to calculate area under PR curve, look for points 61 | # where X axis (recall) changes value 62 | i = np.where(mrec[1:] != mrec[:-1])[0] 63 | 64 | # and sum (\Delta recall) * prec 65 | ap = np.sum((mrec[i + 1] - mrec[i]) * mpre[i + 1]) 66 | return ap 67 | 68 | 69 | def voc_eval(detpath, 70 | annopath, 71 | imagesetfile, 72 | classname, 73 | cachedir, 74 | ovthresh=0.5, 75 | use_07_metric=False): 76 | """rec, prec, ap = voc_eval(detpath, 77 | annopath, 78 | imagesetfile, 79 | classname, 80 | [ovthresh], 81 | [use_07_metric]) 82 | 83 | Top level function that does the PASCAL VOC evaluation. 84 | 85 | detpath: Path to detections 86 | detpath.format(classname) should produce the detection results file. 87 | annopath: Path to annotations 88 | annopath.format(imagename) should be the xml annotations file. 89 | imagesetfile: Text file containing the list of images, one image per line. 90 | classname: Category name (duh) 91 | cachedir: Directory for caching the annotations 92 | [ovthresh]: Overlap threshold (default = 0.5) 93 | [use_07_metric]: Whether to use VOC07's 11 point AP computation 94 | (default False) 95 | """ 96 | # assumes detections are in detpath.format(classname) 97 | # assumes annotations are in annopath.format(imagename) 98 | # assumes imagesetfile is a text file with each line an image name 99 | # cachedir caches the annotations in a pickle file 100 | 101 | # first load gt 102 | if not os.path.isdir(cachedir): 103 | os.mkdir(cachedir) 104 | cachefile = os.path.join(cachedir, '%s_annots.pkl' % imagesetfile.split('/')[-1].split('.txt')[0]) 105 | # read list of images 106 | with open(imagesetfile, 'r') as f: 107 | lines = f.readlines() 108 | imagenames = [x.strip() for x in lines] 109 | 110 | if not os.path.isfile(cachefile): 111 | # load annotations 112 | recs = {} 113 | for i, imagename in enumerate(imagenames): 114 | recs[imagename] = parse_rec(annopath.format(imagename)) 115 | # if i % 100 == 0: 116 | # print('Reading annotation for {:d}/{:d}'.format( 117 | # i + 1, len(imagenames))) 118 | # save 119 | print('Saving cached annotations to {:s}'.format(cachefile)) 120 | with open(cachefile, 'wb') as f: 121 | pickle.dump(recs, f) 122 | else: 123 | # load 124 | with open(cachefile, 'rb') as f: 125 | try: 126 | recs = pickle.load(f) 127 | except: 128 | recs = pickle.load(f, encoding='bytes') 129 | 130 | # extract gt objects for this class 131 | class_recs = {} 132 | npos = 0 133 | for imagename in imagenames: 134 | R = [obj for obj in recs[imagename] if obj['name'] == classname] 135 | bbox = np.array([x['bbox'] for x in R]) 136 | difficult = np.array([x['difficult'] for x in R]).astype(np.bool) 137 | det = [False] * len(R) 138 | npos = npos + sum(~difficult) 139 | class_recs[imagename] = {'bbox': bbox, 140 | 'difficult': difficult, 141 | 'det': det} 142 | 143 | # read dets 144 | detfile = detpath.format(classname) 145 | with open(detfile, 'r') as f: 146 | lines = f.readlines() 147 | 148 | splitlines = [x.strip().split(' ') for x in lines] 149 | image_ids = [x[0] for x in splitlines] 150 | confidence = np.array([float(x[1]) for x in splitlines]) 151 | BB = np.array([[float(z) for z in x[2:]] for x in splitlines]) 152 | 153 | nd = len(image_ids) 154 | tp = np.zeros(nd) 155 | fp = np.zeros(nd) 156 | 157 | if BB.shape[0] > 0: 158 | # sort by confidence 159 | sorted_ind = np.argsort(-confidence) 160 | sorted_scores = np.sort(-confidence) 161 | BB = BB[sorted_ind, :] 162 | image_ids = [image_ids[x] for x in sorted_ind] 163 | 164 | # go down dets and mark TPs and FPs 165 | for d in range(nd): 166 | R = class_recs[image_ids[d]] 167 | bb = BB[d, :].astype(float) 168 | ovmax = -np.inf 169 | BBGT = R['bbox'].astype(float) 170 | 171 | if BBGT.size > 0: 172 | # compute overlaps 173 | # intersection 174 | ixmin = np.maximum(BBGT[:, 0], bb[0]) 175 | iymin = np.maximum(BBGT[:, 1], bb[1]) 176 | ixmax = np.minimum(BBGT[:, 2], bb[2]) 177 | iymax = np.minimum(BBGT[:, 3], bb[3]) 178 | iw = np.maximum(ixmax - ixmin + 1., 0.) 179 | ih = np.maximum(iymax - iymin + 1., 0.) 180 | inters = iw * ih 181 | 182 | # union 183 | uni = ((bb[2] - bb[0] + 1.) * (bb[3] - bb[1] + 1.) + 184 | (BBGT[:, 2] - BBGT[:, 0] + 1.) * 185 | (BBGT[:, 3] - BBGT[:, 1] + 1.) - inters) 186 | 187 | overlaps = inters / uni 188 | ovmax = np.max(overlaps) 189 | jmax = np.argmax(overlaps) 190 | 191 | if ovmax > ovthresh: 192 | if not R['difficult'][jmax]: 193 | if not R['det'][jmax]: 194 | tp[d] = 1. 195 | R['det'][jmax] = 1 196 | else: 197 | fp[d] = 1. 198 | else: 199 | fp[d] = 1. 200 | 201 | # compute precision recall 202 | fp = np.cumsum(fp) 203 | tp = np.cumsum(tp) 204 | rec = tp / float(npos) 205 | # avoid divide by zero in case the first detection matches a difficult 206 | # ground truth 207 | prec = tp / np.maximum(tp + fp, np.finfo(np.float64).eps) 208 | ap = voc_ap(rec, prec, use_07_metric) 209 | 210 | return rec, prec, ap 211 | -------------------------------------------------------------------------------- /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/vg_eval.py: -------------------------------------------------------------------------------- 1 | from __future__ import absolute_import 2 | # -------------------------------------------------------- 3 | # Fast/er R-CNN 4 | # Licensed under The MIT License [see LICENSE for details] 5 | # Written by Bharath Hariharan 6 | # -------------------------------------------------------- 7 | 8 | import xml.etree.ElementTree as ET 9 | import os 10 | import numpy as np 11 | from .voc_eval import voc_ap 12 | 13 | def vg_eval( detpath, 14 | gt_roidb, 15 | image_index, 16 | classindex, 17 | ovthresh=0.5, 18 | use_07_metric=False, 19 | eval_attributes=False): 20 | """rec, prec, ap, sorted_scores, npos = voc_eval( 21 | detpath, 22 | gt_roidb, 23 | image_index, 24 | classindex, 25 | [ovthresh], 26 | [use_07_metric]) 27 | 28 | Top level function that does the Visual Genome evaluation. 29 | 30 | detpath: Path to detections 31 | gt_roidb: List of ground truth structs. 32 | image_index: List of image ids. 33 | classindex: Category index 34 | [ovthresh]: Overlap threshold (default = 0.5) 35 | [use_07_metric]: Whether to use VOC07's 11 point AP computation 36 | (default False) 37 | """ 38 | # extract gt objects for this class 39 | class_recs = {} 40 | npos = 0 41 | for item,imagename in zip(gt_roidb,image_index): 42 | if eval_attributes: 43 | bbox = item['boxes'][np.where(np.any(item['gt_attributes'].toarray() == classindex, axis=1))[0], :] 44 | else: 45 | bbox = item['boxes'][np.where(item['gt_classes'] == classindex)[0], :] 46 | difficult = np.zeros((bbox.shape[0],)).astype(np.bool) 47 | det = [False] * bbox.shape[0] 48 | npos = npos + sum(~difficult) 49 | class_recs[str(imagename)] = {'bbox': bbox, 50 | 'difficult': difficult, 51 | 'det': det} 52 | if npos == 0: 53 | # No ground truth examples 54 | return 0,0,0,0,npos 55 | 56 | # read dets 57 | with open(detpath, 'r') as f: 58 | lines = f.readlines() 59 | if len(lines) == 0: 60 | # No detection examples 61 | return 0,0,0,0,npos 62 | 63 | splitlines = [x.strip().split(' ') for x in lines] 64 | image_ids = [x[0] for x in splitlines] 65 | confidence = np.array([float(x[1]) for x in splitlines]) 66 | BB = np.array([[float(z) for z in x[2:]] for x in splitlines]) 67 | 68 | # sort by confidence 69 | sorted_ind = np.argsort(-confidence) 70 | sorted_scores = -np.sort(-confidence) 71 | BB = BB[sorted_ind, :] 72 | image_ids = [image_ids[x] for x in sorted_ind] 73 | 74 | # go down dets and mark TPs and FPs 75 | nd = len(image_ids) 76 | tp = np.zeros(nd) 77 | fp = np.zeros(nd) 78 | for d in range(nd): 79 | R = class_recs[image_ids[d]] 80 | bb = BB[d, :].astype(float) 81 | ovmax = -np.inf 82 | BBGT = R['bbox'].astype(float) 83 | 84 | if BBGT.size > 0: 85 | # compute overlaps 86 | # intersection 87 | ixmin = np.maximum(BBGT[:, 0], bb[0]) 88 | iymin = np.maximum(BBGT[:, 1], bb[1]) 89 | ixmax = np.minimum(BBGT[:, 2], bb[2]) 90 | iymax = np.minimum(BBGT[:, 3], bb[3]) 91 | iw = np.maximum(ixmax - ixmin + 1., 0.) 92 | ih = np.maximum(iymax - iymin + 1., 0.) 93 | inters = iw * ih 94 | 95 | # union 96 | uni = ((bb[2] - bb[0] + 1.) * (bb[3] - bb[1] + 1.) + 97 | (BBGT[:, 2] - BBGT[:, 0] + 1.) * 98 | (BBGT[:, 3] - BBGT[:, 1] + 1.) - inters) 99 | 100 | overlaps = inters / uni 101 | ovmax = np.max(overlaps) 102 | jmax = np.argmax(overlaps) 103 | 104 | if ovmax > ovthresh: 105 | if not R['difficult'][jmax]: 106 | if not R['det'][jmax]: 107 | tp[d] = 1. 108 | R['det'][jmax] = 1 109 | else: 110 | fp[d] = 1. 111 | else: 112 | fp[d] = 1. 113 | 114 | # compute precision recall 115 | fp = np.cumsum(fp) 116 | tp = np.cumsum(tp) 117 | rec = tp / float(npos) 118 | # avoid divide by zero in case the first detection matches a difficult 119 | # ground truth 120 | prec = tp / np.maximum(tp + fp, np.finfo(np.float64).eps) 121 | ap = voc_ap(rec, prec, use_07_metric) 122 | 123 | return rec, prec, ap, sorted_scores, npos 124 | -------------------------------------------------------------------------------- /lib/datasets/voc_eval.py: -------------------------------------------------------------------------------- 1 | # -------------------------------------------------------- 2 | # Fast/er R-CNN 3 | # Licensed under The MIT License [see LICENSE for details] 4 | # Written by Bharath Hariharan 5 | # -------------------------------------------------------- 6 | from __future__ import absolute_import 7 | from __future__ import division 8 | from __future__ import print_function 9 | 10 | import xml.etree.ElementTree as ET 11 | import os 12 | import pickle 13 | import numpy as np 14 | 15 | def parse_rec(filename): 16 | """ Parse a PASCAL VOC xml file """ 17 | tree = ET.parse(filename) 18 | objects = [] 19 | for obj in tree.findall('object'): 20 | obj_struct = {} 21 | obj_struct['name'] = obj.find('name').text 22 | obj_struct['pose'] = obj.find('pose').text 23 | obj_struct['truncated'] = int(obj.find('truncated').text) 24 | obj_struct['difficult'] = int(obj.find('difficult').text) 25 | bbox = obj.find('bndbox') 26 | obj_struct['bbox'] = [int(bbox.find('xmin').text), 27 | int(bbox.find('ymin').text), 28 | int(bbox.find('xmax').text), 29 | int(bbox.find('ymax').text)] 30 | objects.append(obj_struct) 31 | 32 | return objects 33 | 34 | 35 | def voc_ap(rec, prec, use_07_metric=False): 36 | """ ap = voc_ap(rec, prec, [use_07_metric]) 37 | Compute VOC AP given precision and recall. 38 | If use_07_metric is true, uses the 39 | VOC 07 11 point method (default:False). 40 | """ 41 | if use_07_metric: 42 | # 11 point metric 43 | ap = 0. 44 | for t in np.arange(0., 1.1, 0.1): 45 | if np.sum(rec >= t) == 0: 46 | p = 0 47 | else: 48 | p = np.max(prec[rec >= t]) 49 | ap = ap + p / 11. 50 | else: 51 | # correct AP calculation 52 | # first append sentinel values at the end 53 | mrec = np.concatenate(([0.], rec, [1.])) 54 | mpre = np.concatenate(([0.], prec, [0.])) 55 | 56 | # compute the precision envelope 57 | for i in range(mpre.size - 1, 0, -1): 58 | mpre[i - 1] = np.maximum(mpre[i - 1], mpre[i]) 59 | 60 | # to calculate area under PR curve, look for points 61 | # where X axis (recall) changes value 62 | i = np.where(mrec[1:] != mrec[:-1])[0] 63 | 64 | # and sum (\Delta recall) * prec 65 | ap = np.sum((mrec[i + 1] - mrec[i]) * mpre[i + 1]) 66 | return ap 67 | 68 | 69 | def voc_eval(detpath, 70 | annopath, 71 | imagesetfile, 72 | classname, 73 | cachedir, 74 | ovthresh=0.5, 75 | use_07_metric=False): 76 | """rec, prec, ap = voc_eval(detpath, 77 | annopath, 78 | imagesetfile, 79 | classname, 80 | [ovthresh], 81 | [use_07_metric]) 82 | 83 | Top level function that does the PASCAL VOC evaluation. 84 | 85 | detpath: Path to detections 86 | detpath.format(classname) should produce the detection results file. 87 | annopath: Path to annotations 88 | annopath.format(imagename) should be the xml annotations file. 89 | imagesetfile: Text file containing the list of images, one image per line. 90 | classname: Category name (duh) 91 | cachedir: Directory for caching the annotations 92 | [ovthresh]: Overlap threshold (default = 0.5) 93 | [use_07_metric]: Whether to use VOC07's 11 point AP computation 94 | (default False) 95 | """ 96 | # assumes detections are in detpath.format(classname) 97 | # assumes annotations are in annopath.format(imagename) 98 | # assumes imagesetfile is a text file with each line an image name 99 | # cachedir caches the annotations in a pickle file 100 | 101 | # first load gt 102 | if not os.path.isdir(cachedir): 103 | os.mkdir(cachedir) 104 | cachefile = os.path.join(cachedir, '%s_annots.pkl' % imagesetfile.split('/')[-1].split('.txt')[0]) 105 | # read list of images 106 | with open(imagesetfile, 'r') as f: 107 | lines = f.readlines() 108 | imagenames = [x.strip() for x in lines] 109 | 110 | if not os.path.isfile(cachefile): 111 | # load annotations 112 | recs = {} 113 | for i, imagename in enumerate(imagenames): 114 | recs[imagename] = parse_rec(annopath.format(imagename)) 115 | # if i % 100 == 0: 116 | # print('Reading annotation for {:d}/{:d}'.format( 117 | # i + 1, len(imagenames))) 118 | # save 119 | print('Saving cached annotations to {:s}'.format(cachefile)) 120 | with open(cachefile, 'wb') as f: 121 | pickle.dump(recs, f) 122 | else: 123 | # load 124 | with open(cachefile, 'rb') as f: 125 | try: 126 | recs = pickle.load(f) 127 | except: 128 | recs = pickle.load(f, encoding='bytes') 129 | 130 | # extract gt objects for this class 131 | class_recs = {} 132 | npos = 0 133 | for imagename in imagenames: 134 | R = [obj for obj in recs[imagename] if obj['name'] == classname] 135 | bbox = np.array([x['bbox'] for x in R]) 136 | difficult = np.array([x['difficult'] for x in R]).astype(np.bool) 137 | det = [False] * len(R) 138 | npos = npos + sum(~difficult) 139 | class_recs[imagename] = {'bbox': bbox, 140 | 'difficult': difficult, 141 | 'det': det} 142 | 143 | # read dets 144 | detfile = detpath.format(classname) 145 | with open(detfile, 'r') as f: 146 | lines = f.readlines() 147 | 148 | splitlines = [x.strip().split(' ') for x in lines] 149 | image_ids = [x[0] for x in splitlines] 150 | confidence = np.array([float(x[1]) for x in splitlines]) 151 | BB = np.array([[float(z) for z in x[2:]] for x in splitlines]) 152 | 153 | nd = len(image_ids) 154 | tp = np.zeros(nd) 155 | fp = np.zeros(nd) 156 | 157 | if BB.shape[0] > 0: 158 | # sort by confidence 159 | sorted_ind = np.argsort(-confidence) 160 | sorted_scores = np.sort(-confidence) 161 | BB = BB[sorted_ind, :] 162 | image_ids = [image_ids[x] for x in sorted_ind] 163 | 164 | # go down dets and mark TPs and FPs 165 | for d in range(nd): 166 | R = class_recs[image_ids[d]] 167 | bb = BB[d, :].astype(float) 168 | ovmax = -np.inf 169 | BBGT = R['bbox'].astype(float) 170 | 171 | if BBGT.size > 0: 172 | # compute overlaps 173 | # intersection 174 | ixmin = np.maximum(BBGT[:, 0], bb[0]) 175 | iymin = np.maximum(BBGT[:, 1], bb[1]) 176 | ixmax = np.minimum(BBGT[:, 2], bb[2]) 177 | iymax = np.minimum(BBGT[:, 3], bb[3]) 178 | iw = np.maximum(ixmax - ixmin + 1., 0.) 179 | ih = np.maximum(iymax - iymin + 1., 0.) 180 | inters = iw * ih 181 | 182 | # union 183 | uni = ((bb[2] - bb[0] + 1.) * (bb[3] - bb[1] + 1.) + 184 | (BBGT[:, 2] - BBGT[:, 0] + 1.) * 185 | (BBGT[:, 3] - BBGT[:, 1] + 1.) - inters) 186 | 187 | overlaps = inters / uni 188 | ovmax = np.max(overlaps) 189 | jmax = np.argmax(overlaps) 190 | 191 | if ovmax > ovthresh: 192 | if not R['difficult'][jmax]: 193 | if not R['det'][jmax]: 194 | tp[d] = 1. 195 | R['det'][jmax] = 1 196 | else: 197 | fp[d] = 1. 198 | else: 199 | fp[d] = 1. 200 | 201 | # compute precision recall 202 | fp = np.cumsum(fp) 203 | tp = np.cumsum(tp) 204 | rec = tp / float(npos) 205 | # avoid divide by zero in case the first detection matches a difficult 206 | # ground truth 207 | prec = tp / np.maximum(tp + fp, np.finfo(np.float64).eps) 208 | ap = voc_ap(rec, prec, use_07_metric) 209 | 210 | return rec, prec, ap 211 | -------------------------------------------------------------------------------- /lib/datasets/widerface_eval.py: -------------------------------------------------------------------------------- 1 | # -------------------------------------------------------- 2 | # Fast/er R-CNN 3 | # Licensed under The MIT License [see LICENSE for details] 4 | # Written by Bharath Hariharan 5 | # -------------------------------------------------------- 6 | from __future__ import absolute_import 7 | from __future__ import division 8 | from __future__ import print_function 9 | 10 | import xml.etree.ElementTree as ET 11 | import os 12 | import pickle 13 | import numpy as np 14 | 15 | def parse_rec(filename): 16 | """ Parse a PASCAL VOC xml file """ 17 | tree = ET.parse(filename) 18 | objects = [] 19 | for obj in tree.findall('object'): 20 | obj_struct = {} 21 | obj_struct['name'] = obj.find('name').text 22 | # obj_struct['pose'] = obj.find('pose').text 23 | # obj_struct['truncated'] = int(obj.find('truncated').text) 24 | # obj_struct['difficult'] = int(obj.find('difficult').text) 25 | bbox = obj.find('bndbox') 26 | obj_struct['bbox'] = [int(bbox.find('xmin').text), 27 | int(bbox.find('ymin').text), 28 | int(bbox.find('xmax').text), 29 | int(bbox.find('ymax').text)] 30 | objects.append(obj_struct) 31 | 32 | return objects 33 | 34 | 35 | def voc_ap(rec, prec, use_07_metric=False): 36 | """ ap = voc_ap(rec, prec, [use_07_metric]) 37 | Compute VOC AP given precision and recall. 38 | If use_07_metric is true, uses the 39 | VOC 07 11 point method (default:False). 40 | """ 41 | if use_07_metric: 42 | # 11 point metric 43 | ap = 0. 44 | for t in np.arange(0., 1.1, 0.1): 45 | if np.sum(rec >= t) == 0: 46 | p = 0 47 | else: 48 | p = np.max(prec[rec >= t]) 49 | ap = ap + p / 11. 50 | else: 51 | # correct AP calculation 52 | # first append sentinel values at the end 53 | mrec = np.concatenate(([0.], rec, [1.])) 54 | mpre = np.concatenate(([0.], prec, [0.])) 55 | 56 | # compute the precision envelope 57 | for i in range(mpre.size - 1, 0, -1): 58 | mpre[i - 1] = np.maximum(mpre[i - 1], mpre[i]) 59 | 60 | # to calculate area under PR curve, look for points 61 | # where X axis (recall) changes value 62 | i = np.where(mrec[1:] != mrec[:-1])[0] 63 | 64 | # and sum (\Delta recall) * prec 65 | ap = np.sum((mrec[i + 1] - mrec[i]) * mpre[i + 1]) 66 | return ap 67 | 68 | 69 | def voc_eval(detpath, 70 | annopath, 71 | imagesetfile, 72 | classname, 73 | cachedir, 74 | ovthresh=0.5, 75 | use_07_metric=False): 76 | """rec, prec, ap = voc_eval(detpath, 77 | annopath, 78 | imagesetfile, 79 | classname, 80 | [ovthresh], 81 | [use_07_metric]) 82 | 83 | Top level function that does the PASCAL VOC evaluation. 84 | 85 | detpath: Path to detections 86 | detpath.format(classname) should produce the detection results file. 87 | annopath: Path to annotations 88 | annopath.format(imagename) should be the xml annotations file. 89 | imagesetfile: Text file containing the list of images, one image per line. 90 | classname: Category name (duh) 91 | cachedir: Directory for caching the annotations 92 | [ovthresh]: Overlap threshold (default = 0.5) 93 | [use_07_metric]: Whether to use VOC07's 11 point AP computation 94 | (default False) 95 | """ 96 | # assumes detections are in detpath.format(classname) 97 | # assumes annotations are in annopath.format(imagename) 98 | # assumes imagesetfile is a text file with each line an image name 99 | # cachedir caches the annotations in a pickle file 100 | 101 | # first load gt 102 | if not os.path.isdir(cachedir): 103 | os.mkdir(cachedir) 104 | cachefile = os.path.join(cachedir, '%s_annots.pkl' % imagesetfile.split('/')[-1].split('.txt')[0]) 105 | # read list of images 106 | with open(imagesetfile, 'r') as f: 107 | lines = f.readlines() 108 | imagenames = [x.strip() for x in lines] 109 | 110 | if not os.path.isfile(cachefile): 111 | # load annotations 112 | recs = {} 113 | for i, imagename in enumerate(imagenames): 114 | recs[imagename] = parse_rec(annopath.format(imagename)) 115 | # if i % 100 == 0: 116 | # print('Reading annotation for {:d}/{:d}'.format( 117 | # i + 1, len(imagenames))) 118 | # save 119 | print('Saving cached annotations to {:s}'.format(cachefile)) 120 | with open(cachefile, 'wb') as f: 121 | pickle.dump(recs, f) 122 | else: 123 | # load 124 | with open(cachefile, 'rb') as f: 125 | try: 126 | recs = pickle.load(f) 127 | except: 128 | recs = pickle.load(f, encoding='bytes') 129 | 130 | # extract gt objects for this class 131 | class_recs = {} 132 | npos = 0 133 | for imagename in imagenames: 134 | R = [obj for obj in recs[imagename] if obj['name'] == classname] 135 | bbox = np.array([x['bbox'] for x in R]) 136 | #difficult = np.array([x['difficult'] for x in R]).astype(np.bool) 137 | det = [False] * len(R) 138 | npos = npos + bbox.shape[0] 139 | class_recs[imagename] = {'bbox': bbox, 140 | #'difficult': difficult, 141 | 'det': det} 142 | 143 | # read dets 144 | detfile = detpath.format(classname) 145 | with open(detfile, 'r') as f: 146 | lines = f.readlines() 147 | 148 | splitlines = [x.strip().split(' ') for x in lines] 149 | image_ids = [x[0] for x in splitlines] 150 | confidence = np.array([float(x[1]) for x in splitlines]) 151 | BB = np.array([[float(z) for z in x[2:]] for x in splitlines]) 152 | 153 | nd = len(image_ids) 154 | tp = np.zeros(nd) 155 | fp = np.zeros(nd) 156 | 157 | if BB.shape[0] > 0: 158 | # sort by confidence 159 | sorted_ind = np.argsort(-confidence) 160 | sorted_scores = np.sort(-confidence) 161 | BB = BB[sorted_ind, :] 162 | image_ids = [image_ids[x] for x in sorted_ind] 163 | 164 | # go down dets and mark TPs and FPs 165 | for d in range(nd): 166 | R = class_recs[image_ids[d]] 167 | bb = BB[d, :].astype(float) 168 | ovmax = -np.inf 169 | BBGT = R['bbox'].astype(float) 170 | 171 | if BBGT.size > 0: 172 | # compute overlaps 173 | # intersection 174 | ixmin = np.maximum(BBGT[:, 0], bb[0]) 175 | iymin = np.maximum(BBGT[:, 1], bb[1]) 176 | ixmax = np.minimum(BBGT[:, 2], bb[2]) 177 | iymax = np.minimum(BBGT[:, 3], bb[3]) 178 | iw = np.maximum(ixmax - ixmin + 1., 0.) 179 | ih = np.maximum(iymax - iymin + 1., 0.) 180 | inters = iw * ih 181 | 182 | # union 183 | uni = ((bb[2] - bb[0] + 1.) * (bb[3] - bb[1] + 1.) + 184 | (BBGT[:, 2] - BBGT[:, 0] + 1.) * 185 | (BBGT[:, 3] - BBGT[:, 1] + 1.) - inters) 186 | 187 | overlaps = inters / uni 188 | ovmax = np.max(overlaps) 189 | jmax = np.argmax(overlaps) 190 | 191 | if ovmax > ovthresh: 192 | #if not R['difficult'][jmax]: 193 | if not R['det'][jmax]: 194 | tp[d] = 1. 195 | R['det'][jmax] = 1 196 | else: 197 | fp[d] = 1. 198 | else: 199 | fp[d] = 1. 200 | 201 | # compute precision recall 202 | fp = np.cumsum(fp) 203 | tp = np.cumsum(tp) 204 | rec = tp / float(npos) 205 | # avoid divide by zero in case the first detection matches a difficult 206 | # ground truth 207 | prec = tp / np.maximum(tp + fp, np.finfo(np.float64).eps) 208 | ap = voc_ap(rec, prec, use_07_metric) 209 | 210 | return rec, prec, ap -------------------------------------------------------------------------------- /lib/make.sh: -------------------------------------------------------------------------------- 1 | #!/usr/bin/env bash 2 | 3 | #CUDA_PATH=/usr/local/cuda-8.0${CUDA_PATH:+:${CUDA_PATH}} 4 | #CUDA_PATH=/usr/local/cuda-9.0${CUDA_PATH:+:${CUDA_PATH}} 5 | #PATH=/usr/local/cuda-8.0/bin${PATH:+:${PATH}} 6 | #PATH=/usr/local/cuda-9.0/bin${PATH:+:${PATH}} 7 | #PATH=/usr/local/cuda/include${PATH:+:${PATH}} 8 | 9 | python setup.py build_ext --inplace 10 | rm -rf build 11 | 12 | CUDA_ARCH="-gencode arch=compute_30,code=sm_30 \ 13 | -gencode arch=compute_35,code=sm_35 \ 14 | -gencode arch=compute_50,code=sm_50 \ 15 | -gencode arch=compute_52,code=sm_52 \ 16 | -gencode arch=compute_60,code=sm_60 \ 17 | -gencode arch=compute_61,code=sm_61" 18 | #-gencode arch=compute_70,code=sm_70" 19 | 20 | # compile NMS 21 | cd model/nms/src 22 | echo "Compiling nms kernels by nvcc..." 23 | nvcc -c -o nms_cuda_kernel.cu.o nms_cuda_kernel.cu \ 24 | -D GOOGLE_CUDA=1 -x cu -Xcompiler -fPIC $CUDA_ARCH 25 | 26 | cd ../ 27 | python build.py 28 | 29 | # compile roi_pooling 30 | cd ../../ 31 | cd model/roi_pooling/src 32 | echo "Compiling roi pooling kernels by nvcc..." 33 | nvcc -c -o roi_pooling.cu.o roi_pooling_kernel.cu \ 34 | -D GOOGLE_CUDA=1 -x cu -Xcompiler -fPIC $CUDA_ARCH 35 | cd ../ 36 | python build.py 37 | 38 | # compile roi_align 39 | cd ../../ 40 | cd model/roi_align/src 41 | echo "Compiling roi align kernels by nvcc..." 42 | nvcc -c -o roi_align_kernel.cu.o roi_align_kernel.cu \ 43 | -D GOOGLE_CUDA=1 -x cu -Xcompiler -fPIC $CUDA_ARCH 44 | cd ../ 45 | python build.py 46 | 47 | # compile roi_crop 48 | cd ../../ 49 | cd model/roi_crop/src 50 | echo "Compiling roi crop kernels by nvcc..." 51 | nvcc -c -o roi_crop_cuda_kernel.cu.o roi_crop_cuda_kernel.cu \ 52 | -D GOOGLE_CUDA=1 -x cu -Xcompiler -fPIC $CUDA_ARCH 53 | cd ../ 54 | python build.py 55 | -------------------------------------------------------------------------------- /lib/model/__init__.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/frank-xwang/towards-universal-object-detection/cf40aed4c79b86b3e8e08e4adf94f43742693111/lib/model/__init__.py -------------------------------------------------------------------------------- /lib/model/faster_rcnn/__init__.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/frank-xwang/towards-universal-object-detection/cf40aed4c79b86b3e8e08e4adf94f43742693111/lib/model/faster_rcnn/__init__.py -------------------------------------------------------------------------------- /lib/model/faster_rcnn/domain_attention_module.py: -------------------------------------------------------------------------------- 1 | # 104_x_787 and 105, 112, 111, 130, 152, 140 use it 2 | import torch 3 | import numpy as np 4 | from torch import nn 5 | import torch.nn.functional as F 6 | from model.utils.config import cfg 7 | import torch 8 | from model.faster_rcnn.se_module_vector import SELayer 9 | 10 | class DomainAttention(nn.Module): 11 | def __init__(self, planes, reduction=16, nclass_list=None, fixed_block=False): 12 | super(DomainAttention, self).__init__() 13 | self.planes = planes 14 | num_adapters = cfg.num_adapters 15 | if num_adapters == 0: 16 | self.n_datasets = len(nclass_list) 17 | else: 18 | self.n_datasets = num_adapters 19 | self.fixed_block = fixed_block 20 | self.avg_pool = nn.AdaptiveAvgPool2d(1) 21 | if not self.fixed_block and cfg.less_blocks: 22 | if cfg.block_id != 4: 23 | if cfg.layer_index % 2 == 0: 24 | self.fixed_block = True 25 | else: 26 | if cfg.layer_index % 2 != 0: 27 | self.fixed_block = True 28 | if self.fixed_block or num_adapters == 1: 29 | self.SE_Layers = nn.ModuleList([SELayer(planes, reduction, with_sigmoid=False) for num_class in range(1)]) 30 | elif num_adapters == 0: 31 | self.SE_Layers = nn.ModuleList([SELayer(planes, reduction, with_sigmoid=False) for num_class in nclass_list]) 32 | else: 33 | self.SE_Layers = nn.ModuleList([SELayer(planes, reduction, with_sigmoid=False) for num_class in range(num_adapters)]) 34 | self.fc_1 = nn.Linear(planes, self.n_datasets) 35 | self.sigmoid = nn.Sigmoid() 36 | self.softmax = nn.Softmax(dim=1) 37 | 38 | def forward(self, x): 39 | b, c, _, _ = x.size() 40 | 41 | if self.fixed_block: 42 | SELayers_Matrix = self.SE_Layers[0](x).view(b, c, 1, 1) 43 | SELayers_Matrix = self.sigmoid(SELayers_Matrix) 44 | else: 45 | weight = self.fc_1(self.avg_pool(x).view(b, c)) 46 | weight = self.softmax(weight).view(b, self.n_datasets, 1) 47 | for i, SE_Layer in enumerate(self.SE_Layers): 48 | if i == 0: 49 | SELayers_Matrix = SE_Layer(x).view(b, c, 1) 50 | else: 51 | SELayers_Matrix = torch.cat((SELayers_Matrix, SE_Layer(x).view(b, c, 1)), 2) 52 | SELayers_Matrix = torch.matmul(SELayers_Matrix, weight).view(b, c, 1, 1) 53 | SELayers_Matrix = self.sigmoid(SELayers_Matrix) 54 | return x*SELayers_Matrix -------------------------------------------------------------------------------- /lib/model/faster_rcnn/faster_rcnn.py: -------------------------------------------------------------------------------- 1 | import random 2 | import torch 3 | import torch.nn as nn 4 | import torch.nn.functional as F 5 | from torch.autograd import Variable 6 | import torchvision.models as models 7 | from torch.autograd import Variable 8 | import numpy as np 9 | from model.utils.config import cfg 10 | from model.rpn.rpn import _RPN 11 | from model.roi_pooling.modules.roi_pool import _RoIPooling 12 | from model.roi_crop.modules.roi_crop import _RoICrop 13 | from model.roi_align.modules.roi_align import RoIAlignAvg 14 | from model.rpn.proposal_target_layer_cascade import _ProposalTargetLayer 15 | import time 16 | import pdb 17 | from model.utils.net_utils import _smooth_l1_loss, _crop_pool_layer, _affine_grid_gen, _affine_theta 18 | 19 | class _fasterRCNN(nn.Module): 20 | """ faster RCNN """ 21 | def __init__(self, classes, class_agnostic, rpn_batchsize): 22 | super(_fasterRCNN, self).__init__() 23 | self.classes = classes 24 | self.n_classes = len(classes) 25 | self.class_agnostic = class_agnostic 26 | self.rpn_batchsize = rpn_batchsize 27 | # loss 28 | self.RCNN_loss_cls = 0 29 | self.RCNN_loss_bbox = 0 30 | 31 | # define rpn 32 | self.RCNN_rpn = _RPN(self.dout_base_model, self.rpn_batchsize) 33 | self.RCNN_proposal_target = _ProposalTargetLayer(self.n_classes) 34 | print 'INFO: pooling size is: ', cfg.POOLING_SIZE_H, cfg.POOLING_SIZE_W 35 | self.RCNN_roi_pool = _RoIPooling(cfg.POOLING_SIZE_H, cfg.POOLING_SIZE_W, 1.0/16.0) 36 | self.RCNN_roi_align = RoIAlignAvg(cfg.POOLING_SIZE_H, cfg.POOLING_SIZE_W, 1.0/16.0) 37 | ## wrote by Xudong Wang 38 | self.grid_size_H = cfg.POOLING_SIZE_H * 2 if cfg.CROP_RESIZE_WITH_MAX_POOL else cfg.POOLING_SIZE_H 39 | self.grid_size_W = cfg.POOLING_SIZE_W * 2 if cfg.CROP_RESIZE_WITH_MAX_POOL else cfg.POOLING_SIZE_W 40 | ## end 41 | self.grid_size = cfg.POOLING_SIZE * 2 if cfg.CROP_RESIZE_WITH_MAX_POOL else cfg.POOLING_SIZE 42 | 43 | self.RCNN_roi_crop = _RoICrop() 44 | 45 | def forward(self, im_data, im_info, gt_boxes, num_boxes): 46 | batch_size = im_data.size(0) 47 | im_info = im_info.data 48 | gt_boxes = gt_boxes.data 49 | num_boxes = num_boxes.data 50 | #print('original gt_boxes is ',gt_boxes) 51 | # feed image data to base model to obtain base feature map 52 | cfg.n = 0 53 | x = im_data 54 | base_feat = self.RCNN_base(x) 55 | 56 | # feed base feature map tp RPN to obtain rois 57 | rois, rpn_loss_cls, rpn_loss_bbox = self.RCNN_rpn(base_feat, im_info, gt_boxes, num_boxes) 58 | 59 | # if it is training phrase, then use ground trubut bboxes for refining 60 | if self.training: 61 | roi_data = self.RCNN_proposal_target(rois, gt_boxes, num_boxes) 62 | rois, rois_label, rois_target, rois_inside_ws, rois_outside_ws = roi_data 63 | 64 | rois_label = Variable(rois_label.view(-1).long()) 65 | rois_target = Variable(rois_target.view(-1, rois_target.size(2))) 66 | rois_inside_ws = Variable(rois_inside_ws.view(-1, rois_inside_ws.size(2))) 67 | rois_outside_ws = Variable(rois_outside_ws.view(-1, rois_outside_ws.size(2))) 68 | else: 69 | rois_label = None 70 | rois_target = None 71 | rois_inside_ws = None 72 | rois_outside_ws = None 73 | rpn_loss_cls = 0 74 | rpn_loss_bbox = 0 75 | 76 | rois = Variable(rois) 77 | # do roi pooling based on predicted rois 78 | 79 | if cfg.POOLING_MODE == 'crop': 80 | # pdb.set_trace() 81 | # pooled_feat_anchor = _crop_pool_layer(base_feat, rois.view(-1, 5)) 82 | grid_xy = _affine_grid_gen(rois.view(-1, 5), base_feat.size()[2:], self.grid_size) 83 | grid_yx = torch.stack([grid_xy.data[:,:,:,1], grid_xy.data[:,:,:,0]], 3).contiguous() 84 | pooled_feat = self.RCNN_roi_crop(base_feat, Variable(grid_yx).detach()) 85 | if cfg.CROP_RESIZE_WITH_MAX_POOL: 86 | pooled_feat = F.max_pool2d(pooled_feat, 2, 2) 87 | elif cfg.POOLING_MODE == 'align': 88 | pooled_feat = self.RCNN_roi_align(base_feat, rois.view(-1, 5)) 89 | elif cfg.POOLING_MODE == 'pool': 90 | pooled_feat = self.RCNN_roi_pool(base_feat, rois.view(-1,5)) 91 | 92 | # feed pooled features to top model 93 | pooled_feat = self._head_to_tail(pooled_feat) 94 | 95 | # compute bbox offset 96 | bbox_pred = self.RCNN_bbox_pred(pooled_feat) 97 | if self.training and not self.class_agnostic: 98 | # select the corresponding columns according to roi labels 99 | bbox_pred_view = bbox_pred.view(bbox_pred.size(0), int(bbox_pred.size(1) / 4), 4) 100 | bbox_pred_select = torch.gather(bbox_pred_view, 1, rois_label.view(rois_label.size(0), 1, 1).expand(rois_label.size(0), 1, 4)) 101 | bbox_pred = bbox_pred_select.squeeze(1) 102 | 103 | # compute object classification probability 104 | cls_score = self.RCNN_cls_score(pooled_feat) 105 | cls_prob = F.softmax(cls_score,dim=1) 106 | 107 | RCNN_loss_cls = 0 108 | RCNN_loss_bbox = 0 109 | 110 | if self.training: 111 | # classification loss 112 | RCNN_loss_cls = F.cross_entropy(cls_score, rois_label) 113 | 114 | # bounding box regression L1 loss 115 | RCNN_loss_bbox = _smooth_l1_loss(bbox_pred, rois_target, rois_inside_ws, rois_outside_ws) 116 | 117 | cls_prob = cls_prob.view(batch_size, rois.size(1), -1) 118 | bbox_pred = bbox_pred.view(batch_size, rois.size(1), -1) 119 | 120 | if self.training: 121 | rpn_loss_cls = torch.unsqueeze(rpn_loss_cls, 0) 122 | rpn_loss_bbox = torch.unsqueeze(rpn_loss_bbox, 0) 123 | RCNN_loss_cls = torch.unsqueeze(RCNN_loss_cls, 0) 124 | RCNN_loss_bbox = torch.unsqueeze(RCNN_loss_bbox, 0) 125 | 126 | return rois, cls_prob, bbox_pred, rpn_loss_cls, rpn_loss_bbox, RCNN_loss_cls, RCNN_loss_bbox, rois_label 127 | 128 | def _init_weights(self): 129 | def normal_init(m, mean, stddev, truncated=False): 130 | """ 131 | weight initalizer: truncated normal and random normal. 132 | """ 133 | # x is a parameter 134 | if truncated: 135 | m.weight.data.normal_().fmod_(2).mul_(stddev).add_(mean) # not a perfect approximation 136 | else: 137 | m.weight.data.normal_(mean, stddev) 138 | m.bias.data.zero_() 139 | 140 | normal_init(self.RCNN_rpn.RPN_Conv, 0, 0.01, cfg.TRAIN.TRUNCATED) 141 | normal_init(self.RCNN_rpn.RPN_cls_score, 0, 0.01, cfg.TRAIN.TRUNCATED) 142 | normal_init(self.RCNN_rpn.RPN_bbox_pred, 0, 0.01, cfg.TRAIN.TRUNCATED) 143 | normal_init(self.RCNN_cls_score, 0, 0.01, cfg.TRAIN.TRUNCATED) 144 | normal_init(self.RCNN_bbox_pred, 0, 0.001, cfg.TRAIN.TRUNCATED) 145 | 146 | def create_architecture(self): 147 | self._init_modules() 148 | self._init_weights() 149 | -------------------------------------------------------------------------------- /lib/model/faster_rcnn/faster_rcnn_uni.py: -------------------------------------------------------------------------------- 1 | import random 2 | import torch 3 | import torch.nn as nn 4 | import torch.nn.functional as F 5 | from torch.autograd import Variable 6 | import torchvision.models as models 7 | from torch.autograd import Variable 8 | import numpy as np 9 | from model.utils.config import cfg 10 | from model.rpn.rpn_universal import _RPN 11 | from model.roi_pooling.modules.roi_pool import _RoIPooling 12 | from model.roi_crop.modules.roi_crop import _RoICrop 13 | from model.roi_align.modules.roi_align import RoIAlignAvg 14 | from model.rpn.proposal_target_layer_cascade import _ProposalTargetLayer 15 | import time 16 | import pdb 17 | from model.utils.net_utils import _smooth_l1_loss, _crop_pool_layer, _affine_grid_gen, _affine_theta 18 | 19 | class _fasterRCNN(nn.Module): 20 | """ faster RCNN """ 21 | def __init__(self, classes, class_agnostic, rpn_batchsize_list): 22 | super(_fasterRCNN, self).__init__() 23 | self.classes = classes 24 | self.n_classes = len(classes) 25 | self.class_agnostic = class_agnostic 26 | self.rpn_batchsize_list = rpn_batchsize_list 27 | # loss 28 | self.RCNN_loss_cls = 0 29 | self.RCNN_loss_bbox = 0 30 | 31 | # define rpn 32 | self.RCNN_rpn = _RPN(self.dout_base_model, self.rpn_batchsize_list) 33 | self.RCNN_proposal_target = _ProposalTargetLayer(self.n_classes) 34 | self.RCNN_roi_pool = _RoIPooling(cfg.POOLING_SIZE_H, cfg.POOLING_SIZE_W, 1.0/16.0) 35 | self.RCNN_roi_align = RoIAlignAvg(cfg.POOLING_SIZE_H, cfg.POOLING_SIZE_W, 1.0/16.0) 36 | 37 | self.grid_size_H = cfg.POOLING_SIZE_H * 2 if cfg.CROP_RESIZE_WITH_MAX_POOL else cfg.POOLING_SIZE_H 38 | self.grid_size_W = cfg.POOLING_SIZE_W * 2 if cfg.CROP_RESIZE_WITH_MAX_POOL else cfg.POOLING_SIZE_W 39 | ## end 40 | self.grid_size = cfg.POOLING_SIZE * 2 if cfg.CROP_RESIZE_WITH_MAX_POOL else cfg.POOLING_SIZE 41 | 42 | self.RCNN_roi_crop = _RoICrop() 43 | 44 | def forward(self, im_data, im_info, gt_boxes, num_boxes, cls_ind): 45 | batch_size = im_data.size(0) 46 | 47 | im_info = im_info.data 48 | gt_boxes = gt_boxes.data 49 | num_boxes = num_boxes.data 50 | # feed image data to base model to obtain base feature map 51 | cfg.n = 0 52 | x = im_data 53 | x = self.RCNN_base(x) 54 | base_feat = x 55 | 56 | # feed base feature map tp RPN to obtain rois 57 | rois, rpn_loss_cls, rpn_loss_bbox = self.RCNN_rpn(base_feat, im_info, gt_boxes, num_boxes, cfg.cls_ind) 58 | 59 | # if it is training phrase, then use ground trubut bboxes for refining 60 | if self.training: 61 | roi_data = self.RCNN_proposal_target(rois, gt_boxes, num_boxes) 62 | rois, rois_label, rois_target, rois_inside_ws, rois_outside_ws = roi_data 63 | 64 | rois_label = Variable(rois_label.view(-1).long()) 65 | rois_target = Variable(rois_target.view(-1, rois_target.size(2))) 66 | rois_inside_ws = Variable(rois_inside_ws.view(-1, rois_inside_ws.size(2))) 67 | rois_outside_ws = Variable(rois_outside_ws.view(-1, rois_outside_ws.size(2))) 68 | else: 69 | rois_label = None 70 | rois_target = None 71 | rois_inside_ws = None 72 | rois_outside_ws = None 73 | rpn_loss_cls = 0 74 | rpn_loss_bbox = 0 75 | 76 | rois = Variable(rois) 77 | # do roi pooling based on predicted rois 78 | if cfg.POOLING_MODE == 'crop': 79 | # pdb.set_trace() 80 | grid_xy = _affine_grid_gen(rois.view(-1, 5), base_feat.size()[2:], self.grid_size) 81 | grid_yx = torch.stack([grid_xy.data[:,:,:,1], grid_xy.data[:,:,:,0]], 3).contiguous() 82 | pooled_feat = self.RCNN_roi_crop(base_feat, Variable(grid_yx).detach()) 83 | if cfg.CROP_RESIZE_WITH_MAX_POOL: 84 | pooled_feat = F.max_pool2d(pooled_feat, 2, 2) 85 | elif cfg.POOLING_MODE == 'align': 86 | pooled_feat = self.RCNN_roi_align(base_feat, rois.view(-1, 5)) 87 | elif cfg.POOLING_MODE == 'pool': 88 | pooled_feat = self.RCNN_roi_pool(base_feat, rois.view(-1,5)) 89 | # feed pooled features to top model 90 | pooled_feat = self._head_to_tail(pooled_feat) 91 | bbox_pred = self.RCNN_bbox_pred_layers[cfg.cls_ind](pooled_feat) 92 | if self.training and not self.class_agnostic: 93 | # select the corresponding columns according to roi labels 94 | bbox_pred_view = bbox_pred.view(bbox_pred.size(0), int(bbox_pred.size(1) / 4), 4) 95 | bbox_pred_select = torch.gather(bbox_pred_view, 1, rois_label.view(rois_label.size(0), 1, 1).expand(rois_label.size(0), 1, 4)) 96 | bbox_pred = bbox_pred_select.squeeze(1) 97 | # compute object classification probability 98 | cls_score = self.RCNN_cls_score_layers[cfg.cls_ind](pooled_feat) 99 | cls_prob = F.softmax(cls_score,dim=1) 100 | 101 | RCNN_loss_cls = 0 102 | RCNN_loss_bbox = 0 103 | 104 | if self.training: 105 | # classification loss 106 | RCNN_loss_cls = F.cross_entropy(cls_score, rois_label) 107 | # bounding box regression L1 loss 108 | RCNN_loss_bbox = _smooth_l1_loss(bbox_pred, rois_target, rois_inside_ws, rois_outside_ws) 109 | 110 | cls_prob = cls_prob.view(batch_size, rois.size(1), -1) 111 | bbox_pred = bbox_pred.view(batch_size, rois.size(1), -1) 112 | 113 | if self.training: 114 | rpn_loss_cls = torch.unsqueeze(rpn_loss_cls, 0) 115 | rpn_loss_bbox = torch.unsqueeze(rpn_loss_bbox, 0) 116 | RCNN_loss_cls = torch.unsqueeze(RCNN_loss_cls, 0) 117 | RCNN_loss_bbox = torch.unsqueeze(RCNN_loss_bbox, 0) 118 | 119 | return rois, cls_prob, bbox_pred, rpn_loss_cls, rpn_loss_bbox, RCNN_loss_cls, RCNN_loss_bbox, rois_label 120 | 121 | def _init_weights(self): 122 | def normal_init(m, mean, stddev, truncated=False): 123 | """ 124 | weight initalizer: truncated normal and random normal. 125 | """ 126 | # x is a parameter 127 | if truncated: 128 | m.weight.data.normal_().fmod_(2).mul_(stddev).add_(mean) # not a perfect approximation 129 | else: 130 | m.weight.data.normal_(mean, stddev) 131 | m.bias.data.zero_() 132 | for i in range(len(self.RCNN_rpn.RPN_cls_score_layers)): 133 | normal_init(self.RCNN_rpn.RPN_cls_score_layers[i], 0, 0.01, cfg.TRAIN.TRUNCATED) 134 | normal_init(self.RCNN_rpn.RPN_bbox_pred_layers[i], 0, 0.01, cfg.TRAIN.TRUNCATED) 135 | normal_init(self.RCNN_cls_score_layers[i], 0, 0.01, cfg.TRAIN.TRUNCATED) 136 | normal_init(self.RCNN_bbox_pred_layers[i], 0, 0.001, cfg.TRAIN.TRUNCATED) 137 | 138 | def create_architecture(self): 139 | self._init_modules() 140 | self._init_weights() 141 | -------------------------------------------------------------------------------- /lib/model/faster_rcnn/se_module.py: -------------------------------------------------------------------------------- 1 | from torch import nn 2 | import torch.nn.functional as F 3 | 4 | class SELayer(nn.Module): 5 | def __init__(self, channel, reduction=16): 6 | super(SELayer, self).__init__() 7 | self.avg_pool = nn.AdaptiveAvgPool2d(1) 8 | self.fc = nn.Sequential( 9 | nn.Linear(channel, channel // reduction), 10 | nn.ReLU(inplace=True), 11 | nn.Linear(channel // reduction, channel), 12 | nn.Sigmoid() 13 | ) 14 | 15 | def forward(self, x): 16 | b, c, _, _ = x.size() 17 | y = self.avg_pool(x).view(b, c) 18 | y = self.fc(y).view(b, c, 1, 1) 19 | return x * y -------------------------------------------------------------------------------- /lib/model/faster_rcnn/se_module_vector.py: -------------------------------------------------------------------------------- 1 | from torch import nn 2 | import torch.nn.functional as F 3 | 4 | class SELayer(nn.Module): 5 | def __init__(self, channel, reduction=16, with_sigmoid=True): 6 | super(SELayer, self).__init__() 7 | self.with_sigmoid = with_sigmoid 8 | self.avg_pool = nn.AdaptiveAvgPool2d(1) 9 | if with_sigmoid: 10 | self.fc = nn.Sequential( 11 | nn.Linear(channel, channel // reduction), 12 | nn.ReLU(inplace=True), 13 | nn.Linear(channel // reduction, channel), 14 | nn.Sigmoid() 15 | ) 16 | else: 17 | self.fc = nn.Sequential( 18 | nn.Linear(channel, channel // reduction), 19 | nn.ReLU(inplace=True), 20 | nn.Linear(channel // reduction, channel), 21 | ) 22 | 23 | def forward(self, x): 24 | b, c, _, _ = x.size() 25 | y = self.avg_pool(x).view(b, c) 26 | y = self.fc(y).view(b, c, 1, 1) 27 | return y -------------------------------------------------------------------------------- /lib/model/nms/.gitignore: -------------------------------------------------------------------------------- 1 | *.c 2 | *.cpp 3 | *.so 4 | -------------------------------------------------------------------------------- /lib/model/nms/__init__.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/frank-xwang/towards-universal-object-detection/cf40aed4c79b86b3e8e08e4adf94f43742693111/lib/model/nms/__init__.py -------------------------------------------------------------------------------- /lib/model/nms/_ext/__init__.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/frank-xwang/towards-universal-object-detection/cf40aed4c79b86b3e8e08e4adf94f43742693111/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.maximum(x2[i], x2[order[1:]]) 24 | yy2 = np.maximum(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 | 36 | 37 | -------------------------------------------------------------------------------- /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.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/frank-xwang/towards-universal-object-detection/cf40aed4c79b86b3e8e08e4adf94f43742693111/lib/model/roi_align/__init__.py -------------------------------------------------------------------------------- /lib/model/roi_align/_ext/__init__.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/frank-xwang/towards-universal-object-detection/cf40aed4c79b86b3e8e08e4adf94f43742693111/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/frank-xwang/towards-universal-object-detection/cf40aed4c79b86b3e8e08e4adf94f43742693111/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/frank-xwang/towards-universal-object-detection/cf40aed4c79b86b3e8e08e4adf94f43742693111/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.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.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/frank-xwang/towards-universal-object-detection/cf40aed4c79b86b3e8e08e4adf94f43742693111/lib/model/roi_crop/__init__.py -------------------------------------------------------------------------------- /lib/model/roi_crop/_ext/__init__.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/frank-xwang/towards-universal-object-detection/cf40aed4c79b86b3e8e08e4adf94f43742693111/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/frank-xwang/towards-universal-object-detection/cf40aed4c79b86b3e8e08e4adf94f43742693111/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/frank-xwang/towards-universal-object-detection/cf40aed4c79b86b3e8e08e4adf94f43742693111/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/frank-xwang/towards-universal-object-detection/cf40aed4c79b86b3e8e08e4adf94f43742693111/lib/model/roi_pooling/__init__.py -------------------------------------------------------------------------------- /lib/model/roi_pooling/_ext/__init__.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/frank-xwang/towards-universal-object-detection/cf40aed4c79b86b3e8e08e4adf94f43742693111/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/frank-xwang/towards-universal-object-detection/cf40aed4c79b86b3e8e08e4adf94f43742693111/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/frank-xwang/towards-universal-object-detection/cf40aed4c79b86b3e8e08e4adf94f43742693111/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.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/frank-xwang/towards-universal-object-detection/cf40aed4c79b86b3e8e08e4adf94f43742693111/lib/model/rpn/__init__.py -------------------------------------------------------------------------------- /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 | def generate_anchors(base_size=16, ratios=[0.5, 1, 2], 45 | scales=2**np.arange(3, 6)): 46 | """ 47 | Generate anchor (reference) windows by enumerating aspect ratios X 48 | scales wrt a reference (0, 0, 15, 15) window. 49 | """ 50 | 51 | base_anchor = np.array([1, 1, base_size, base_size]) - 1 52 | #print('base anchor is: ', base_anchor) 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 | #print('all anchors are: ',anchors) 57 | return anchors 58 | 59 | def _whctrs(anchor): 60 | """ 61 | Return width, height, x center, and y center for an anchor (window). 62 | """ 63 | 64 | w = anchor[2] - anchor[0] + 1 65 | h = anchor[3] - anchor[1] + 1 66 | x_ctr = anchor[0] + 0.5 * (w - 1) 67 | y_ctr = anchor[1] + 0.5 * (h - 1) 68 | return w, h, x_ctr, y_ctr 69 | 70 | def _mkanchors(ws, hs, x_ctr, y_ctr): 71 | """ 72 | Given a vector of widths (ws) and heights (hs) around a center 73 | (x_ctr, y_ctr), output a set of anchors (windows). 74 | """ 75 | 76 | ws = ws[:, np.newaxis] 77 | hs = hs[:, np.newaxis] 78 | anchors = np.hstack((x_ctr - 0.5 * (ws - 1), 79 | y_ctr - 0.5 * (hs - 1), 80 | x_ctr + 0.5 * (ws - 1), 81 | y_ctr + 0.5 * (hs - 1))) 82 | return anchors 83 | 84 | def _ratio_enum(anchor, ratios): 85 | """ 86 | Enumerate a set of anchors for each aspect ratio wrt an anchor. 87 | """ 88 | 89 | w, h, x_ctr, y_ctr = _whctrs(anchor) 90 | #print(w, h, x_ctr, y_ctr) 91 | size = w * h 92 | size_ratios = size / ratios 93 | #print('size_ratios',size_ratios) 94 | ws = np.round(np.sqrt(size_ratios)) 95 | hs = np.round(ws * ratios) 96 | #print(ws, hs, x_ctr, y_ctr) 97 | anchors = _mkanchors(ws, hs, x_ctr, y_ctr) 98 | #print('base anchors with ratio is:', anchors) 99 | return anchors 100 | 101 | def _scale_enum(anchor, scales): 102 | """ 103 | Enumerate a set of anchors for each scale wrt an anchor. 104 | """ 105 | w, h, x_ctr, y_ctr = _whctrs(anchor) 106 | #print('scaled: ',w, h, x_ctr, y_ctr) 107 | ws = w * scales 108 | hs = h * scales 109 | #print('width and height are: ',ws,hs) 110 | anchors = _mkanchors(ws, hs, x_ctr, y_ctr) 111 | return anchors 112 | 113 | if __name__ == '__main__': 114 | import time 115 | t = time.time() 116 | a = generate_anchors() 117 | print(time.time() - t) 118 | print(a) 119 | from IPython import embed; embed() -------------------------------------------------------------------------------- /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, rpn_batchsize): 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 | self.rpn_batchsize = rpn_batchsize 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,self.rpn_batchsize) 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, ): 59 | 60 | batch_size = base_feat.size(0) 61 | 62 | # return feature map after convrelu layer 63 | #print('base_feat size: ', base_feat.size()) 64 | rpn_conv1 = F.relu(self.RPN_Conv(base_feat), inplace=True) 65 | # get rpn classification score 66 | rpn_cls_score = self.RPN_cls_score(rpn_conv1) 67 | 68 | rpn_cls_score_reshape = self.reshape(rpn_cls_score, 2) 69 | <<<<<<< HEAD 70 | rpn_cls_prob_reshape = F.softmax(rpn_cls_score_reshape, dim=1) 71 | ======= 72 | rpn_cls_prob_reshape = F.softmax(rpn_cls_score_reshape) 73 | >>>>>>> 793eeda709a4483589939795954491531204c768 74 | rpn_cls_prob = self.reshape(rpn_cls_prob_reshape, self.nc_score_out) 75 | 76 | # get rpn offsets to the anchor boxes 77 | rpn_bbox_pred = self.RPN_bbox_pred(rpn_conv1) 78 | 79 | # proposal layer 80 | cfg_key = 'TRAIN' if self.training else 'TEST' 81 | self.rpn_loss_cls = 0 82 | self.rpn_loss_box = 0 83 | # generating training labels and build the rpn loss 84 | if self.training: 85 | assert gt_boxes is not None 86 | if cfg.sample_mode == 'bootstrap': 87 | rpn_data = self.RPN_anchor_target((rpn_cls_score.data, gt_boxes, im_info, num_boxes, rpn_cls_prob.data)) 88 | else: 89 | rpn_data = self.RPN_anchor_target((rpn_cls_score.data, gt_boxes, im_info, num_boxes)) 90 | 91 | # compute classification loss 92 | rpn_cls_score = rpn_cls_score_reshape.permute(0, 2, 3, 1).contiguous().view(batch_size, -1, 2) 93 | rpn_label = rpn_data[0].view(batch_size, -1) 94 | 95 | rpn_keep = Variable(rpn_label.view(-1).ne(-1).nonzero().view(-1)) 96 | rpn_cls_score = torch.index_select(rpn_cls_score.view(-1,2), 0, rpn_keep) 97 | rpn_label = torch.index_select(rpn_label.view(-1), 0, rpn_keep.data) 98 | rpn_label = Variable(rpn_label.long()) 99 | self.rpn_loss_cls = F.cross_entropy(rpn_cls_score, rpn_label) 100 | fg_cnt = torch.sum(rpn_label.data.ne(0)) 101 | 102 | rpn_bbox_targets, rpn_bbox_inside_weights, rpn_bbox_outside_weights = rpn_data[1:] 103 | 104 | # compute bbox regression loss 105 | rpn_bbox_inside_weights = Variable(rpn_bbox_inside_weights) 106 | rpn_bbox_outside_weights = Variable(rpn_bbox_outside_weights) 107 | rpn_bbox_targets = Variable(rpn_bbox_targets) 108 | 109 | self.rpn_loss_box = _smooth_l1_loss(rpn_bbox_pred, rpn_bbox_targets, rpn_bbox_inside_weights, 110 | rpn_bbox_outside_weights, sigma=3, dim=[1,2,3]) 111 | 112 | # In RPN_proposal layer, we will choose 2000 proposals from about 20,000 after this step 113 | # they will sort all proposals w.r.t score(12,000 will be saved) and use nms to choose 114 | # 2000 proposals from the 12,000 proposals 115 | if self.training and (cfg.use_coco_igonore==False): 116 | # rois[:,:,0] is the batch size number 117 | rois = self.RPN_proposal((rpn_cls_prob.data, rpn_bbox_pred.data, 118 | im_info, cfg_key, gt_boxes, num_boxes, rpn_cls_score.data)) 119 | else: 120 | rois = self.RPN_proposal((rpn_cls_prob.data, rpn_bbox_pred.data, 121 | im_info, cfg_key)) 122 | 123 | return rois, self.rpn_loss_cls, self.rpn_loss_box 124 | -------------------------------------------------------------------------------- /lib/model/rpn/rpn_universal.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, rpn_batchsize_list): 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 | self.rpn_batchsize_list = rpn_batchsize_list 27 | 28 | # define the convrelu layers processing input feature map 29 | self.RPN_Conv = nn.Conv2d(self.din, 512, 3, 1, 1, bias=True) 30 | 31 | # define bg/fg classifcation score layer 32 | # self.nc_score_out = len(self.anchor_scales) * len(self.anchor_ratios) * 2 # 2(bg/fg) * 9 (anchors) 33 | self.RPN_cls_score_layers = nn.ModuleList([nn.Conv2d(512, int(2*nc_score_out), 1, 1, 0) for nc_score_out in cfg.ANCHOR_NUM]) 34 | 35 | # define anchor box offset prediction layer 36 | # self.nc_bbox_out = len(self.anchor_scales) * len(self.anchor_ratios) * 4 # 4(coords) * 9 (anchors) 37 | self.RPN_bbox_pred_layers = nn.ModuleList([nn.Conv2d(512, int(4*nc_score_out), 1, 1, 0) for nc_score_out in cfg.ANCHOR_NUM]) 38 | 39 | # define proposal layer 40 | self.RPN_proposal = nn.ModuleList([_ProposalLayer(self.feat_stride, cfg.ANCHOR_SCALES_LIST[i], cfg.ANCHOR_RATIOS_LIST[i]) for i in np.arange(len(cfg.ANCHOR_NUM))]) 41 | 42 | # define anchor target layer 43 | # define module list for diffrent _AnchorTargetLayer 44 | self.RPN_anchor_target_layers = nn.ModuleList([_AnchorTargetLayer(self.feat_stride, cfg.ANCHOR_SCALES_LIST[i], 45 | cfg.ANCHOR_RATIOS_LIST[i], cfg.train_rpn_batchsize_list[i]) for i in np.arange(len(cfg.ANCHOR_NUM))]) 46 | 47 | self.rpn_loss_cls = 0 48 | self.rpn_loss_box = 0 49 | 50 | @staticmethod 51 | def reshape(x, d): 52 | input_shape = x.size() 53 | x = x.view( 54 | input_shape[0], 55 | int(d), 56 | int(float(input_shape[1] * input_shape[2]) / float(d)), 57 | input_shape[3] 58 | ) 59 | return x 60 | 61 | def forward(self, base_feat, im_info, gt_boxes, num_boxes, cls_ind): 62 | if cfg.rpn_univ: 63 | cls_ind = 0 64 | else: 65 | cls_ind = cfg.cls_ind 66 | 67 | batch_size = base_feat.size(0) 68 | 69 | # return feature map after convrelu layer 70 | rpn_conv1 = F.relu(self.RPN_Conv(base_feat), inplace=True) 71 | # get rpn classification score 72 | rpn_cls_score = self.RPN_cls_score_layers[cls_ind](rpn_conv1) 73 | 74 | rpn_cls_score_reshape = self.reshape(rpn_cls_score, 2) 75 | rpn_cls_prob_reshape = F.softmax(rpn_cls_score_reshape, dim=1) 76 | self.nc_score_out = int(cfg.ANCHOR_NUM[cls_ind]*2) 77 | rpn_cls_prob = self.reshape(rpn_cls_prob_reshape, self.nc_score_out) 78 | 79 | # get rpn offsets to the anchor boxes 80 | rpn_bbox_pred = self.RPN_bbox_pred_layers[cls_ind](rpn_conv1) 81 | 82 | # proposal layer 83 | cfg_key = 'TRAIN' if self.training else 'TEST' 84 | 85 | rois = self.RPN_proposal[cls_ind]((rpn_cls_prob.data, rpn_bbox_pred.data, 86 | im_info, cfg_key)) 87 | 88 | self.rpn_loss_cls = 0 89 | self.rpn_loss_box = 0 90 | 91 | # generating training labels and build the rpn loss 92 | if self.training: 93 | assert gt_boxes is not None 94 | 95 | # decide which RPN_anchor_target layers to use 96 | rpn_data = self.RPN_anchor_target_layers[cls_ind]((rpn_cls_score.data, gt_boxes, im_info, num_boxes)) 97 | 98 | # compute classification loss 99 | rpn_cls_score = rpn_cls_score_reshape.permute(0, 2, 3, 1).contiguous().view(batch_size, -1, 2) 100 | rpn_label = rpn_data[0].view(batch_size, -1) 101 | 102 | rpn_keep = Variable(rpn_label.view(-1).ne(-1).nonzero().view(-1)) 103 | rpn_cls_score = torch.index_select(rpn_cls_score.view(-1,2), 0, rpn_keep) 104 | rpn_label = torch.index_select(rpn_label.view(-1), 0, rpn_keep.data) 105 | rpn_label = Variable(rpn_label.long()) 106 | self.rpn_loss_cls = F.cross_entropy(rpn_cls_score, rpn_label) 107 | fg_cnt = torch.sum(rpn_label.data.ne(0)) 108 | 109 | rpn_bbox_targets, rpn_bbox_inside_weights, rpn_bbox_outside_weights = rpn_data[1:] 110 | 111 | # compute bbox regression loss 112 | rpn_bbox_inside_weights = Variable(rpn_bbox_inside_weights) 113 | rpn_bbox_outside_weights = Variable(rpn_bbox_outside_weights) 114 | rpn_bbox_targets = Variable(rpn_bbox_targets) 115 | 116 | self.rpn_loss_box = _smooth_l1_loss(rpn_bbox_pred, rpn_bbox_targets, rpn_bbox_inside_weights, 117 | rpn_bbox_outside_weights, sigma=3, dim=[1,2,3]) 118 | 119 | 120 | return rois, self.rpn_loss_cls, self.rpn_loss_box -------------------------------------------------------------------------------- /lib/model/utils/.gitignore: -------------------------------------------------------------------------------- 1 | *.c 2 | *.cpp 3 | *.so 4 | -------------------------------------------------------------------------------- /lib/model/utils/__init__.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/frank-xwang/towards-universal-object-detection/cf40aed4c79b86b3e8e08e4adf94f43742693111/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 106 | -------------------------------------------------------------------------------- /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 /= 255. # Convert range to [0,1] 40 | # print(np.mean(im), np.std(im)) 41 | # normalization for pytroch pretrained models. 42 | # https://pytorch.org/docs/stable/torchvision/models.html 43 | # Original pixel mean is: [102.9801, 115.9465, 122.7717] 44 | pixel_means = [0.485, 0.456, 0.406] 45 | pixel_stdens = [0.229, 0.224, 0.225] 46 | 47 | # normalize manual 48 | im -= pixel_means # Minus mean 49 | im /= pixel_stdens # divide by stddev 50 | # im -= pixel_means 51 | # im = im[:, :, ::-1] 52 | im_shape = im.shape 53 | im_size_min = np.min(im_shape[0:2]) 54 | im_size_max = np.max(im_shape[0:2]) 55 | im_scale = float(target_size) / float(im_size_min) 56 | # Prevent the biggest axis from being more than MAX_SIZE 57 | # if np.round(im_scale * im_size_max) > max_size: 58 | # im_scale = float(max_size) / float(im_size_max) 59 | # im = imresize(im, im_scale) 60 | #print('im shape brfore: ',im.shape, im_scale, target_size, im_size_min) 61 | im = cv2.resize(im, None, None, fx=im_scale, fy=im_scale, 62 | interpolation=cv2.INTER_LINEAR) 63 | #print('im shape brfore: ',im.shape) 64 | 65 | return im, im_scale 66 | -------------------------------------------------------------------------------- /lib/model/utils/logger.py: -------------------------------------------------------------------------------- 1 | # Code referenced from https://gist.github.com/gyglim/1f8dfb1b5c82627ae3efcfbbadb9f514 2 | import tensorflow as tf 3 | import numpy as np 4 | import scipy.misc 5 | try: 6 | from StringIO import StringIO # Python 2.7 7 | except ImportError: 8 | from io import BytesIO # Python 3.x 9 | 10 | 11 | class Logger(object): 12 | 13 | def __init__(self, log_dir): 14 | """Create a summary writer logging to log_dir.""" 15 | self.writer = tf.summary.FileWriter(log_dir) 16 | 17 | def scalar_summary(self, tag, value, step): 18 | """Log a scalar variable.""" 19 | summary = tf.Summary(value=[tf.Summary.Value(tag=tag, simple_value=value)]) 20 | self.writer.add_summary(summary, step) 21 | 22 | def image_summary(self, tag, images, step): 23 | """Log a list of images.""" 24 | 25 | img_summaries = [] 26 | for i, img in enumerate(images): 27 | # Write the image to a string 28 | try: 29 | s = StringIO() 30 | except: 31 | s = BytesIO() 32 | scipy.misc.toimage(img).save(s, format="png") 33 | 34 | # Create an Image object 35 | img_sum = tf.Summary.Image(encoded_image_string=s.getvalue(), 36 | height=img.shape[0], 37 | width=img.shape[1]) 38 | # Create a Summary value 39 | img_summaries.append(tf.Summary.Value(tag='%s/%d' % (tag, i), image=img_sum)) 40 | 41 | # Create and write Summary 42 | summary = tf.Summary(value=img_summaries) 43 | self.writer.add_summary(summary, step) 44 | 45 | def histo_summary(self, tag, values, step, bins=1000): 46 | """Log a histogram of the tensor of values.""" 47 | 48 | # Create a histogram using numpy 49 | counts, bin_edges = np.histogram(values, bins=bins) 50 | 51 | # Fill the fields of the histogram proto 52 | hist = tf.HistogramProto() 53 | hist.min = float(np.min(values)) 54 | hist.max = float(np.max(values)) 55 | hist.num = int(np.prod(values.shape)) 56 | hist.sum = float(np.sum(values)) 57 | hist.sum_squares = float(np.sum(values**2)) 58 | 59 | # Drop the start of the first bin 60 | bin_edges = bin_edges[1:] 61 | 62 | # Add bin edges and counts 63 | for edge in bin_edges: 64 | hist.bucket_limit.append(edge) 65 | for c in counts: 66 | hist.bucket.append(c) 67 | 68 | # Create and write Summary 69 | summary = tf.Summary(value=[tf.Summary.Value(tag=tag, histo=hist)]) 70 | self.writer.add_summary(summary, step) 71 | self.writer.flush() 72 | -------------------------------------------------------------------------------- /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.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.blob import prep_im_for_blob, im_list_to_blob 17 | import pdb 18 | from model.utils.config import cfg 19 | 20 | def get_minibatch(roidb, num_classes, target_size576=1, target_size600=1, target_size800=1): 21 | """Given a roidb, construct a minibatch sampled from it.""" 22 | num_images = len(roidb) 23 | # Sample random scales to use for each image in this batch 24 | random_scale_inds = npr.randint(0, high=len(cfg.TRAIN.SCALES), 25 | size=num_images) 26 | # print('train scales is: ', cfg.TRAIN.SCALES, num_classes) 27 | assert(cfg.TRAIN.BATCH_SIZE % num_images == 0), \ 28 | 'num_images ({}) must divide BATCH_SIZE ({})'. \ 29 | format(num_images, cfg.TRAIN.BATCH_SIZE) 30 | 31 | # Get the input image blob, formatted for caffe 32 | im_blob, im_scales = _get_image_blob(roidb, random_scale_inds, target_size576, target_size600, target_size800) 33 | 34 | blobs = {'data': im_blob} 35 | 36 | assert len(im_scales) == 1, "Single batch only" 37 | assert len(roidb) == 1, "Single batch only" 38 | 39 | # gt boxes: (x1, y1, x2, y2, cls) 40 | if cfg.TRAIN.USE_ALL_GT: 41 | # Include all ground truth boxes 42 | gt_inds = np.where(roidb[0]['gt_classes'] != 0)[0] 43 | else: 44 | # For the COCO ground truth boxes, exclude the ones that are ''iscrowd'' 45 | gt_inds = np.where(roidb[0]['gt_classes'] != 0 & np.all(roidb[0]['gt_overlaps'].toarray() > -1.0, axis=1))[0] 46 | gt_boxes = np.empty((len(gt_inds), 5), dtype=np.float32) 47 | gt_boxes[:, 0:4] = roidb[0]['boxes'][gt_inds, :] * im_scales[0] 48 | gt_boxes[:, 4] = roidb[0]['gt_classes'][gt_inds] 49 | blobs['gt_boxes'] = gt_boxes 50 | blobs['im_info'] = np.array( 51 | [[im_blob.shape[1], im_blob.shape[2], im_scales[0]]], 52 | dtype=np.float32) 53 | 54 | blobs['img_id'] = roidb[0]['img_id'] 55 | 56 | return blobs 57 | 58 | def _get_image_blob(roidb, scale_inds, target_size576=1,target_size600=1, target_size800=1): 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 | 81 | if cfg.random_resize: 82 | if target_size == 600 and cfg.universal: 83 | target_size = target_size*target_size600 84 | elif target_size >= 780 and cfg.universal: 85 | target_size = target_size*target_size800 86 | elif target_size == 576 and cfg.universal: 87 | target_size = int(target_size576) 88 | #print('target_size',target_size, cfg.size, cfg.TRAIN.MAX_SIZE) 89 | #print('cfg.TRAIN.SCALES', cfg.TRAIN.SCALES) 90 | im, im_scale = prep_im_for_blob(im, cfg.PIXEL_MEANS, target_size, 91 | cfg.TRAIN.MAX_SIZE) 92 | im_scales.append(im_scale) 93 | processed_ims.append(im) 94 | 95 | # Create a blob to hold the input images 96 | blob = im_list_to_blob(processed_ims) 97 | 98 | return blob, im_scales 99 | -------------------------------------------------------------------------------- /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 | import torch 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 | roidb = imdb.roidb 21 | if not (imdb.name.startswith('coco')): 22 | sizes = [PIL.Image.open(imdb.image_path_at(i)).size 23 | for i in range(imdb.num_images)] 24 | fg_num = 0 25 | for i in range(len(imdb.image_index)): 26 | roidb[i]['img_id'] = imdb.image_id_at(i) 27 | roidb[i]['image'] = imdb.image_path_at(i) 28 | if not (imdb.name.startswith('coco')): 29 | roidb[i]['width'] = sizes[i][0] 30 | roidb[i]['height'] = sizes[i][1] 31 | # need gt_overlaps as a dense array for argmax 32 | gt_overlaps = roidb[i]['gt_overlaps'].toarray() 33 | # max overlap with gt over classes (columns) 34 | max_overlaps = gt_overlaps.max(axis=1) 35 | # gt class that had the max overlap 36 | max_classes = gt_overlaps.argmax(axis=1) 37 | roidb[i]['max_classes'] = max_classes 38 | roidb[i]['max_overlaps'] = max_overlaps 39 | # sanity checks 40 | zero_inds = np.where(max_overlaps == 0)[0] 41 | assert all(max_classes[zero_inds] == 0) 42 | nonzero_inds = np.where(max_overlaps > 0)[0] 43 | fg_num += len(np.where(max_overlaps == 1)[0]) 44 | 45 | def rank_roidb_ratio(roidb,imdb_names): 46 | # rank roidb based on the ratio between width and height. 47 | if 'kitti' in imdb_names: 48 | print('current imdb name is kitti, ratio_large is set as 4(original is 2)') 49 | ratio_large = 4.0 50 | ratio_small = 0.5 51 | else: 52 | ratio_large = 2.0 # largest ratio to preserve. 53 | ratio_small = 0.5 # smallest ratio to preserve. 54 | 55 | ratio_list = [] 56 | for i in range(len(roidb)): 57 | width = roidb[i]['width'] 58 | height = roidb[i]['height'] 59 | ratio = width / float(height) 60 | #print('ratio is: ',ratio, width, height) 61 | 62 | if ratio > ratio_large: 63 | roidb[i]['need_crop'] = 1 64 | ratio = ratio_large 65 | elif ratio < ratio_small: 66 | roidb[i]['need_crop'] = 1 67 | ratio = ratio_small 68 | else: 69 | roidb[i]['need_crop'] = 0 70 | 71 | ratio_list.append(ratio) 72 | 73 | ratio_list = np.array(ratio_list) 74 | ratio_index = np.argsort(ratio_list) 75 | return ratio_list[ratio_index], ratio_index 76 | 77 | def filter_roidb(roidb): 78 | # filter the image without bounding box. 79 | print('before filtering, contains %d images...' % (len(roidb))) 80 | i = 0 81 | while i < len(roidb): 82 | if cfg.imdb_name == "KITTI": 83 | roidb[i]['boxes'] +=1 84 | if (len(roidb[i]['boxes']) == 0): 85 | if cfg.filter_empty: 86 | del roidb[i] 87 | i -= 1 88 | else: 89 | i+=1 90 | continue 91 | max_num = int(np.max(roidb[i]['boxes'])) 92 | min_num = int(np.min(roidb[i]['boxes'])) 93 | min_y1 = int(np.min(roidb[i]['boxes'][:,1])) 94 | max_y2 = int(np.max(roidb[i]['boxes'][:,3])) 95 | min_x1 = int(np.min(roidb[i]['boxes'][:,0])) 96 | max_x2 = int(np.max(roidb[i]['boxes'][:,2])) 97 | if max_num > 60000: 98 | print('error, please check rois coordinate of: \n %s is correct'%(roidb[i]['image'])) 99 | print(roidb[i]['boxes']) 100 | index = np.where(roidb[i]['boxes'] > 60000) 101 | roidb[i]['boxes'][index] = 0 102 | print(roidb[i]['boxes']) 103 | if min_x1 == max_x2 or min_y1 == max_y2: 104 | del roidb[i] 105 | i -= 1 106 | continue 107 | i+=1 108 | 109 | print('after filtering, contains %d images...' % (len(roidb))) 110 | return roidb 111 | 112 | def combined_roidb(imdb_names, training=True): 113 | """ 114 | Combine multiple roidbs 115 | """ 116 | 117 | def get_training_roidb(imdb): 118 | """Returns a roidb (Region of Interest database) for use in training.""" 119 | if cfg.TRAIN.USE_FLIPPED: 120 | print('Appending horizontally-flipped training examples...') 121 | imdb.append_flipped_images() 122 | print('done') 123 | 124 | print('Preparing roidb data...') 125 | 126 | prepare_roidb(imdb) 127 | print('done') 128 | 129 | return imdb.roidb 130 | 131 | def get_roidb(imdb_name): 132 | imdb = get_imdb(imdb_name) 133 | print('dataset `{:s}` is loaded'.format(imdb.name)) 134 | imdb.set_proposal_method(cfg.TRAIN.PROPOSAL_METHOD) 135 | roidb = get_training_roidb(imdb) 136 | return roidb 137 | roidbs = [get_roidb(s) for s in imdb_names.split('+')] 138 | roidb = roidbs[0] 139 | 140 | if len(roidbs) > 1: 141 | for r in roidbs[1:]: 142 | roidb.extend(r) 143 | tmp = get_imdb(imdb_names.split('+')[1]) 144 | imdb = datasets.imdb.imdb(imdb_names, tmp.classes) 145 | else: 146 | imdb = get_imdb(imdb_names) 147 | 148 | if training: 149 | roidb = filter_roidb(roidb) 150 | ratio_list, ratio_index = rank_roidb_ratio(roidb, imdb_names) 151 | return imdb, roidb, ratio_list, ratio_index -------------------------------------------------------------------------------- /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 | -------------------------------------------------------------------------------- /lib/training_info.py: -------------------------------------------------------------------------------- 1 | def train_params(net, aim): 2 | if aim == 'fine_tune': 3 | if net == 'vgg16_bn_finetune': 4 | #number of epochs to train 5 | PRE_datasets=pascal_voc_0712 6 | PRE_epoch=11 7 | PRE_ses=10 8 | PRE_checkpoint=8274 9 | #datasets=caltech 10 | datasets=KITTI 11 | #datasets=pascal_voc_0712 12 | #datasets=coco 13 | #datasets=universal 14 | datasets=widerface 15 | 16 | return BATCH_SIZE, checkepoch, SESSION, DECAY_STEP, 17 | if net == 'res18_fine_tune': 18 | BATCH_SIZE=4 19 | WORKER_NUMBER=8 20 | LEARNING_RATE=0.004 21 | DECAY_STEP=8 22 | GPU_ID=4,5,6,8 23 | GPU_ID=5,6,7,8 24 | GPU_ID=2,7 25 | checkepoch=2 26 | SESSION=12 27 | CHECKPOINT=6437 28 | #number of epochs to train 29 | epochs=14 30 | resume=True 31 | backward_together=False 32 | PRE_datasets=pascal_voc_0712 33 | PRE_epoch=11 34 | PRE_ses=10 35 | PRE_checkpoint=8274 36 | #datasets=caltech 37 | datasets=KITTI 38 | #datasets=pascal_voc_0712 39 | #datasets=coco 40 | #datasets=universal 41 | datasets=widerface -------------------------------------------------------------------------------- /requirements.txt: -------------------------------------------------------------------------------- 1 | Cython 2 | cffi 3 | opencv-python 4 | scipy==1.1.0 5 | msgpack 6 | easydict 7 | matplotlib 8 | pyyaml 9 | tensorboardX 10 | Pillow 11 | -------------------------------------------------------------------------------- /scripts/test_universal.sh: -------------------------------------------------------------------------------- 1 | GPU_ID=0 2 | batch_size=1 3 | net=da_res50 4 | DATA_DIR=data 5 | num_adapters=8 6 | less_blocks=True # False 7 | 8 | ### Arguments for checkpoints 9 | EPOCH=14 10 | SESSION=8 11 | CHECKPOINT=13331 12 | 13 | ### Choose the datasest to test 14 | datasets_test=KITTI 15 | # datasets_list='KITTI widerface pascal_voc_0712 Kitchen LISA' 16 | datasets_list='LISA pascal_voc_0712 Kitchen coco clipart watercolor comic widerface dota deeplesion KITTI' 17 | 18 | CUDA_DEVICE_ORDER=PCI_BUS_ID CUDA_VISIBLE_DEVICES=${GPU_ID} \ 19 | python test_universal.py \ 20 | --dataset ${datasets_test} --net ${net} \ 21 | --checksession ${SESSION} \ 22 | --checkepoch ${EPOCH} \ 23 | --checkpoint ${CHECKPOINT} \ 24 | --cuda --mGPUs \ 25 | --DATA_DIR ${DATA_DIR} \ 26 | --num_adapters ${num_adapters} \ 27 | --less_blocks ${less_blocks} \ 28 | --datasets_list ${datasets_list} \ 29 | --bs ${batch_size} 30 | -------------------------------------------------------------------------------- /scripts/train_universal.sh: -------------------------------------------------------------------------------- 1 | GPU_ID=0,1,2,3,4,5,6,7 2 | BATCH_SIZE=16 3 | net=da_res50 4 | WORKER_NUMBER=2 5 | LEARNING_RATE=0.01 # lr=0.01 for BATCH_SIZE=16 6 | DECAY_STEP=10 7 | SAVE_SESSION=11 8 | epochs=14 9 | backward_together=0 # 0: independent; 1: together 10 | USE_FLIPPED=1 # choose 1 for using flipped images, 0 for don't 11 | datasets=universal 12 | DATA_DIR=data 13 | random_resize=True 14 | fix_bn=True 15 | use_bn_mux=False 16 | update_chosen=False 17 | randomly_chosen_datasets=True 18 | warmup_steps=0 19 | num_adapters=11 20 | less_blocks=False 21 | datasets_list='KITTI widerface pascal_voc_0712 Kitchen LISA deeplesion coco clipart comic watercolor dota' 22 | # datasets_list='KITTI widerface pascal_voc_0712 Kitchen LISA' 23 | # resume=True 24 | # checkepoch=7 25 | # checksession=11083 26 | # CHECKPOINT=5720 27 | 28 | CUDA_VISIBLE_DEVICES=${GPU_ID} python universal_model.py \ 29 | --dataset ${datasets} --net ${net} \ 30 | --bs ${BATCH_SIZE} --nw ${WORKER_NUMBER} \ 31 | --lr ${LEARNING_RATE} --lr_decay_step ${DECAY_STEP} \ 32 | --USE_FLIPPED ${USE_FLIPPED} \ 33 | --cuda --mGPUs \ 34 | --s ${SAVE_SESSION} \ 35 | --epochs ${epochs} \ 36 | --DATA_DIR ${DATA_DIR} \ 37 | --random_resize ${random_resize} \ 38 | --num_adapters ${num_adapters} \ 39 | --less_blocks ${less_blocks} \ 40 | --fix_bn ${fix_bn} \ 41 | --use_mux ${use_bn_mux} \ 42 | --randomly_chosen_datasets ${randomly_chosen_datasets} \ 43 | --update_chosen ${update_chosen} \ 44 | --warmup_steps ${warmup_steps} \ 45 | --backward_together ${backward_together} \ 46 | --datasets_list ${datasets_list} 47 | # --r ${resume} \ 48 | # --checksession ${checksession} \ 49 | # --checkpoint ${CHECKPOINT} \ 50 | # --checkepoch ${checkepoch} \ --------------------------------------------------------------------------------