├── .gitignore ├── ABSTRACTIONS.md ├── CODE_OF_CONDUCT.md ├── CONTRIBUTING.md ├── INSTALL.md ├── LICENSE ├── README.md ├── TROUBLESHOOTING.md ├── configs ├── pytorch_mask_rcnn_R2_101_s4_FPN_2x.yaml ├── pytorch_mask_rcnn_R2_50_s2_FPN_2x.yaml ├── pytorch_mask_rcnn_R2_50_s4_FPN_2x.yaml ├── pytorch_mask_rcnn_R2_50_s6_FPN_2x.yaml ├── pytorch_mask_rcnn_R2_50_s8_FPN_2x.yaml ├── pytorch_mask_rcnn_R_101_FPN_2x.yaml └── pytorch_mask_rcnn_R_50_FPN_2x.yaml ├── demo ├── README.md ├── demo_e2e_mask_rcnn_R_50_FPN_1x.png ├── demo_e2e_mask_rcnn_X_101_32x8d_FPN_1x.png ├── predictor.py └── webcam.py ├── docker ├── Dockerfile └── docker-jupyter │ ├── Dockerfile │ └── jupyter_notebook_config.py ├── inference └── coco_2014_minival │ ├── bbox.json │ └── segm.json ├── maskrcnn_benchmark ├── __init__.py ├── config │ ├── __init__.py │ ├── defaults.py │ └── paths_catalog.py ├── csrc │ ├── ROIAlign.h │ ├── ROIPool.h │ ├── SigmoidFocalLoss.h │ ├── cpu │ │ ├── ROIAlign_cpu.cpp │ │ ├── nms_cpu.cpp │ │ └── vision.h │ ├── cuda │ │ ├── ROIAlign_cuda.cu │ │ ├── ROIPool_cuda.cu │ │ ├── SigmoidFocalLoss_cuda.cu │ │ ├── nms.cu │ │ └── vision.h │ ├── nms.h │ └── vision.cpp ├── data │ ├── README.md │ ├── __init__.py │ ├── build.py │ ├── collate_batch.py │ ├── datasets │ │ ├── __init__.py │ │ ├── coco.py │ │ ├── concat_dataset.py │ │ ├── evaluation │ │ │ ├── __init__.py │ │ │ ├── coco │ │ │ │ ├── __init__.py │ │ │ │ └── coco_eval.py │ │ │ └── voc │ │ │ │ ├── __init__.py │ │ │ │ └── voc_eval.py │ │ ├── list_dataset.py │ │ └── voc.py │ ├── samplers │ │ ├── __init__.py │ │ ├── distributed.py │ │ ├── grouped_batch_sampler.py │ │ └── iteration_based_batch_sampler.py │ └── transforms │ │ ├── __init__.py │ │ ├── build.py │ │ └── transforms.py ├── engine │ ├── __init__.py │ ├── inference.py │ └── trainer.py ├── layers │ ├── __init__.py │ ├── _utils.py │ ├── batch_norm.py │ ├── misc.py │ ├── nms.py │ ├── roi_align.py │ ├── roi_pool.py │ ├── sigmoid_focal_loss.py │ └── smooth_l1_loss.py ├── modeling │ ├── __init__.py │ ├── backbone │ │ ├── __init__.py │ │ ├── backbone.py │ │ ├── fbnet.py │ │ ├── fbnet_builder.py │ │ ├── fbnet_modeldef.py │ │ ├── fpn.py │ │ ├── res2net.py │ │ ├── res2net_builder.py │ │ └── resnet.py │ ├── balanced_positive_negative_sampler.py │ ├── box_coder.py │ ├── detector │ │ ├── __init__.py │ │ ├── detectors.py │ │ └── generalized_rcnn.py │ ├── make_layers.py │ ├── matcher.py │ ├── poolers.py │ ├── registry.py │ ├── roi_heads │ │ ├── __init__.py │ │ ├── box_head │ │ │ ├── __init__.py │ │ │ ├── box_head.py │ │ │ ├── inference.py │ │ │ ├── loss.py │ │ │ ├── roi_box_feature_extractors.py │ │ │ └── roi_box_predictors.py │ │ ├── keypoint_head │ │ │ ├── __init__.py │ │ │ ├── inference.py │ │ │ ├── keypoint_head.py │ │ │ ├── loss.py │ │ │ ├── roi_keypoint_feature_extractors.py │ │ │ └── roi_keypoint_predictors.py │ │ ├── mask_head │ │ │ ├── __init__.py │ │ │ ├── inference.py │ │ │ ├── loss.py │ │ │ ├── mask_head.py │ │ │ ├── roi_mask_feature_extractors.py │ │ │ └── roi_mask_predictors.py │ │ └── roi_heads.py │ ├── rpn │ │ ├── __init__.py │ │ ├── anchor_generator.py │ │ ├── inference.py │ │ ├── loss.py │ │ ├── retinanet │ │ │ ├── __init__.py │ │ │ ├── inference.py │ │ │ ├── loss.py │ │ │ └── retinanet.py │ │ ├── rpn.py │ │ └── utils.py │ └── utils.py ├── solver │ ├── __init__.py │ ├── build.py │ └── lr_scheduler.py ├── structures │ ├── __init__.py │ ├── bounding_box.py │ ├── boxlist_ops.py │ ├── image_list.py │ ├── keypoint.py │ └── segmentation_mask.py └── utils │ ├── README.md │ ├── __init__.py │ ├── c2_model_loading.py │ ├── checkpoint.py │ ├── collect_env.py │ ├── comm.py │ ├── cv2_util.py │ ├── env.py │ ├── imports.py │ ├── logger.py │ ├── metric_logger.py │ ├── miscellaneous.py │ ├── model_serialization.py │ ├── model_zoo.py │ └── registry.py ├── setup.py ├── tests ├── checkpoint.py ├── env_tests │ └── env.py ├── test_backbones.py ├── test_box_coder.py ├── test_configs.py ├── test_data_samplers.py ├── test_fbnet.py ├── test_feature_extractors.py ├── test_metric_logger.py ├── test_nms.py ├── test_predictors.py ├── test_rpn_heads.py └── utils.py └── tools ├── cityscapes ├── convert_cityscapes_to_coco.py └── instances2dict_with_polygons.py ├── test_net.py └── train_net.py /.gitignore: -------------------------------------------------------------------------------- 1 | # compilation and distribution 2 | __pycache__ 3 | _ext 4 | *.pyc 5 | *.so 6 | maskrcnn_benchmark.egg-info/ 7 | build/ 8 | dist/ 9 | log.txt 10 | # pytorch/python/numpy formats 11 | *.pth 12 | *.pkl 13 | *.npy 14 | 15 | # ipython/jupyter notebooks 16 | *.ipynb 17 | **/.ipynb_checkpoints/ 18 | 19 | # Editor temporaries 20 | *.swn 21 | *.swo 22 | *.swp 23 | *~ 24 | 25 | # Pycharm editor settings 26 | .idea 27 | 28 | # project dirs 29 | /datasets 30 | /models 31 | -------------------------------------------------------------------------------- /ABSTRACTIONS.md: -------------------------------------------------------------------------------- 1 | ## Abstractions 2 | The main abstractions introduced by `maskrcnn_benchmark` that are useful to 3 | have in mind are the following: 4 | 5 | ### ImageList 6 | In PyTorch, the first dimension of the input to the network generally represents 7 | the batch dimension, and thus all elements of the same batch have the same 8 | height / width. 9 | In order to support images with different sizes and aspect ratios in the same 10 | batch, we created the `ImageList` class, which holds internally a batch of 11 | images (os possibly different sizes). The images are padded with zeros such that 12 | they have the same final size and batched over the first dimension. The original 13 | sizes of the images before padding are stored in the `image_sizes` attribute, 14 | and the batched tensor in `tensors`. 15 | We provide a convenience function `to_image_list` that accepts a few different 16 | input types, including a list of tensors, and returns an `ImageList` object. 17 | 18 | ```python 19 | from maskrcnn_benchmark.structures.image_list import to_image_list 20 | 21 | images = [torch.rand(3, 100, 200), torch.rand(3, 150, 170)] 22 | batched_images = to_image_list(images) 23 | 24 | # it is also possible to make the final batched image be a multiple of a number 25 | batched_images_32 = to_image_list(images, size_divisible=32) 26 | ``` 27 | 28 | ### BoxList 29 | The `BoxList` class holds a set of bounding boxes (represented as a `Nx4` tensor) for 30 | a specific image, as well as the size of the image as a `(width, height)` tuple. 31 | It also contains a set of methods that allow to perform geometric 32 | transformations to the bounding boxes (such as cropping, scaling and flipping). 33 | The class accepts bounding boxes from two different input formats: 34 | - `xyxy`, where each box is encoded as a `x1`, `y1`, `x2` and `y2` coordinates, and 35 | - `xywh`, where each box is encoded as `x1`, `y1`, `w` and `h`. 36 | 37 | Additionally, each `BoxList` instance can also hold arbitrary additional information 38 | for each bounding box, such as labels, visibility, probability scores etc. 39 | 40 | Here is an example on how to create a `BoxList` from a list of coordinates: 41 | ```python 42 | from maskrcnn_benchmark.structures.bounding_box import BoxList, FLIP_LEFT_RIGHT 43 | 44 | width = 100 45 | height = 200 46 | boxes = [ 47 | [0, 10, 50, 50], 48 | [50, 20, 90, 60], 49 | [10, 10, 50, 50] 50 | ] 51 | # create a BoxList with 3 boxes 52 | bbox = BoxList(boxes, image_size=(width, height), mode='xyxy') 53 | 54 | # perform some box transformations, has similar API as PIL.Image 55 | bbox_scaled = bbox.resize((width * 2, height * 3)) 56 | bbox_flipped = bbox.transpose(FLIP_LEFT_RIGHT) 57 | 58 | # add labels for each bbox 59 | labels = torch.tensor([0, 10, 1]) 60 | bbox.add_field('labels', labels) 61 | 62 | # bbox also support a few operations, like indexing 63 | # here, selects boxes 0 and 2 64 | bbox_subset = bbox[[0, 2]] 65 | ``` 66 | -------------------------------------------------------------------------------- /CODE_OF_CONDUCT.md: -------------------------------------------------------------------------------- 1 | # Code of Conduct 2 | 3 | Facebook has adopted a Code of Conduct that we expect project participants to adhere to. 4 | Please read the [full text](https://code.fb.com/codeofconduct/) 5 | so that you can understand what actions will and will not be tolerated. 6 | -------------------------------------------------------------------------------- /CONTRIBUTING.md: -------------------------------------------------------------------------------- 1 | # Contributing to Mask-RCNN Benchmark 2 | We want to make contributing to this project as easy and transparent as 3 | possible. 4 | 5 | ## Our Development Process 6 | Minor changes and improvements will be released on an ongoing basis. Larger changes (e.g., changesets implementing a new paper) will be released on a more periodic basis. 7 | 8 | ## Pull Requests 9 | We actively welcome your pull requests. 10 | 11 | 1. Fork the repo and create your branch from `master`. 12 | 2. If you've added code that should be tested, add tests. 13 | 3. If you've changed APIs, update the documentation. 14 | 4. Ensure the test suite passes. 15 | 5. Make sure your code lints. 16 | 6. If you haven't already, complete the Contributor License Agreement ("CLA"). 17 | 18 | ## Contributor License Agreement ("CLA") 19 | In order to accept your pull request, we need you to submit a CLA. You only need 20 | to do this once to work on any of Facebook's open source projects. 21 | 22 | Complete your CLA here: 23 | 24 | ## Issues 25 | We use GitHub issues to track public bugs. Please ensure your description is 26 | clear and has sufficient instructions to be able to reproduce the issue. 27 | 28 | Facebook has a [bounty program](https://www.facebook.com/whitehat/) for the safe 29 | disclosure of security bugs. In those cases, please go through the process 30 | outlined on that page and do not file a public issue. 31 | 32 | ## Coding Style 33 | * 4 spaces for indentation rather than tabs 34 | * 80 character line length 35 | * PEP8 formatting following [Black](https://black.readthedocs.io/en/stable/) 36 | 37 | ## License 38 | By contributing to Mask-RCNN Benchmark, you agree that your contributions will be licensed 39 | under the LICENSE file in the root directory of this source tree. 40 | -------------------------------------------------------------------------------- /INSTALL.md: -------------------------------------------------------------------------------- 1 | ## Installation 2 | 3 | ### Requirements: 4 | - PyTorch 1.0 from a nightly release. Installation instructions can be found in https://pytorch.org/get-started/locally/ 5 | - torchvision from master 6 | - cocoapi 7 | - yacs 8 | - matplotlib 9 | - GCC >= 4.9 10 | - (optional) OpenCV for the webcam demo 11 | 12 | 13 | ### Option 1: Step-by-step installation 14 | 15 | ```bash 16 | # first, make sure that your conda is setup properly with the right environment 17 | # for that, check that `which conda`, `which pip` and `which python` points to the 18 | # right path. From a clean conda env, this is what you need to do 19 | 20 | conda create --name maskrcnn_benchmark 21 | conda activate maskrcnn_benchmark 22 | 23 | # this installs the right pip and dependencies for the fresh python 24 | conda install ipython 25 | 26 | # maskrcnn_benchmark and coco api dependencies 27 | pip install ninja yacs cython matplotlib 28 | 29 | # follow PyTorch installation in https://pytorch.org/get-started/locally/ 30 | # we give the instructions for CUDA 9.0 31 | conda install pytorch-nightly cudatoolkit=9.0 -c pytorch 32 | 33 | export INSTALL_DIR=$PWD 34 | # install torchvision 35 | cd $INSTALL_DIR 36 | git clone https://github.com/pytorch/vision.git 37 | cd vision 38 | python setup.py install 39 | 40 | # install pycocotools 41 | cd $INSTALL_DIR 42 | git clone https://github.com/cocodataset/cocoapi.git 43 | cd cocoapi/PythonAPI 44 | python setup.py build_ext install 45 | 46 | # install PyTorch Detection 47 | cd $INSTALL_DIR 48 | git clone https://github.com/gasvn/Res2Net-maskrcnn.git 49 | cd Res2Net-maskrcnn 50 | # the following will install the lib with 51 | # symbolic links, so that you can modify 52 | # the files if you want and won't need to 53 | # re-build it 54 | python setup.py build develop 55 | 56 | unset INSTALL_DIR 57 | 58 | # or if you are on macOS 59 | # MACOSX_DEPLOYMENT_TARGET=10.9 CC=clang CXX=clang++ python setup.py build develop 60 | ``` 61 | 62 | ### Option 2: Docker Image (Requires CUDA, Linux only) 63 | 64 | Build image with defaults (`CUDA=9.0`, `CUDNN=7`): 65 | 66 | nvidia-docker build -t maskrcnn-benchmark docker/ 67 | 68 | Build image with other CUDA and CUDNN versions: 69 | 70 | nvidia-docker build -t maskrcnn-benchmark --build-arg CUDA=9.2 --build-arg CUDNN=7 docker/ 71 | 72 | Build and run image with built-in jupyter notebook(note that the password is used to log in jupyter notebook): 73 | 74 | nvidia-docker build -t maskrcnn-benchmark-jupyter docker/docker-jupyter/ 75 | nvidia-docker run -td -p 8888:8888 -e PASSWORD= -v : maskrcnn-benchmark-jupyter 76 | -------------------------------------------------------------------------------- /LICENSE: -------------------------------------------------------------------------------- 1 | MIT License 2 | 3 | Copyright (c) 2018 Facebook 4 | 5 | Permission is hereby granted, free of charge, to any person obtaining a copy 6 | of this software and associated documentation files (the "Software"), to deal 7 | in the Software without restriction, including without limitation the rights 8 | to use, copy, modify, merge, publish, distribute, sublicense, and/or sell 9 | copies of the Software, and to permit persons to whom the Software is 10 | furnished to do so, subject to the following conditions: 11 | 12 | The above copyright notice and this permission notice shall be included in all 13 | copies or substantial portions of the Software. 14 | 15 | THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR 16 | IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 17 | FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE 18 | AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER 19 | LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, 20 | OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE 21 | SOFTWARE. 22 | -------------------------------------------------------------------------------- /TROUBLESHOOTING.md: -------------------------------------------------------------------------------- 1 | # Troubleshooting 2 | 3 | Here is a compilation if common issues that you might face 4 | while compiling / running this code: 5 | 6 | ## Compilation errors when compiling the library 7 | If you encounter build errors like the following: 8 | ``` 9 | /usr/include/c++/6/type_traits:1558:8: note: provided for ‘template struct std::is_convertible’ 10 | struct is_convertible 11 | ^~~~~~~~~~~~~~ 12 | /usr/include/c++/6/tuple:502:1: error: body of constexpr function ‘static constexpr bool std::_TC<, _Elements>::_NonNestedTuple() [with _SrcTuple = std::tuple&&; bool = true; _Elements = {at::Tensor, at::Tensor, at::Tensor, at::Tensor}]’ not a return-statement 13 | } 14 | ^ 15 | error: command '/usr/local/cuda/bin/nvcc' failed with exit status 1 16 | ``` 17 | check your CUDA version and your `gcc` version. 18 | ``` 19 | nvcc --version 20 | gcc --version 21 | ``` 22 | If you are using CUDA 9.0 and gcc 6.4.0, then refer to https://github.com/facebookresearch/maskrcnn-benchmark/issues/25, 23 | which has a summary of the solution. Basically, CUDA 9.0 is not compatible with gcc 6.4.0. 24 | 25 | ## ImportError: No module named maskrcnn_benchmark.config when running webcam.py 26 | 27 | This means that `maskrcnn-benchmark` has not been properly installed. 28 | Refer to https://github.com/facebookresearch/maskrcnn-benchmark/issues/22 for a few possible issues. 29 | Note that we now support Python 2 as well. 30 | 31 | 32 | ## ImportError: Undefined symbol: __cudaPopCallConfiguration error when import _C 33 | 34 | This probably means that the inconsistent version of NVCC compile and your conda CUDAToolKit package. This is firstly mentioned in https://github.com/facebookresearch/maskrcnn-benchmark/issues/45 . All you need to do is: 35 | 36 | ``` 37 | # Check the NVCC compile version(e.g.) 38 | /usr/cuda-9.2/bin/nvcc --version 39 | # Check the CUDAToolKit version(e.g.) 40 | ~/anaconda3/bin/conda list | grep cuda 41 | 42 | # If you need to update your CUDAToolKit 43 | ~/anaconda3/bin/conda install -c anaconda cudatoolkit==9.2 44 | ``` 45 | 46 | Both of them should have the **same** version. For example, if NVCC==9.2 and CUDAToolKit==9.2, this will be fine while when NVCC==9.2 but CUDAToolKit==9, it fails. 47 | 48 | 49 | ## Segmentation fault (core dumped) when running the library 50 | This probably means that you have compiled the library using GCC < 4.9, which is ABI incompatible with PyTorch. 51 | Indeed, during installation, you probably saw a message like 52 | ``` 53 | Your compiler (g++ 4.8) may be ABI-incompatible with PyTorch! 54 | Please use a compiler that is ABI-compatible with GCC 4.9 and above. 55 | See https://gcc.gnu.org/onlinedocs/libstdc++/manual/abi.html. 56 | 57 | See https://gist.github.com/goldsborough/d466f43e8ffc948ff92de7486c5216d6 58 | for instructions on how to install GCC 4.9 or higher. 59 | ``` 60 | Follow the instructions on https://gist.github.com/goldsborough/d466f43e8ffc948ff92de7486c5216d6 61 | to install GCC 4.9 or higher, and try recompiling `maskrcnn-benchmark` again, after cleaning the 62 | `build` folder with 63 | ``` 64 | rm -rf build 65 | ``` 66 | 67 | 68 | -------------------------------------------------------------------------------- /configs/pytorch_mask_rcnn_R2_101_s4_FPN_2x.yaml: -------------------------------------------------------------------------------- 1 | MODEL: 2 | META_ARCHITECTURE: "GeneralizedRCNN" 3 | WEIGHT: "/home/shgao/.torch/models/res2net101_26w_4s-02a759a1.pth" 4 | BACKBONE: 5 | CONV_BODY: "R2-101-FPN" 6 | RESNETS: 7 | BACKBONE_OUT_CHANNELS: 256 8 | WIDTH_PER_GROUP: 26 9 | SCALE: 4 10 | TRANS_FUNC: "Bottle2neckWithFixedBatchNorm" 11 | RPN: 12 | USE_FPN: True 13 | ANCHOR_STRIDE: (4, 8, 16, 32, 64) 14 | PRE_NMS_TOP_N_TRAIN: 2000 15 | PRE_NMS_TOP_N_TEST: 1000 16 | POST_NMS_TOP_N_TEST: 1000 17 | FPN_POST_NMS_TOP_N_TEST: 1000 18 | ROI_HEADS: 19 | USE_FPN: True 20 | ROI_BOX_HEAD: 21 | POOLER_RESOLUTION: 7 22 | POOLER_SCALES: (0.25, 0.125, 0.0625, 0.03125) 23 | POOLER_SAMPLING_RATIO: 2 24 | FEATURE_EXTRACTOR: "FPN2MLPFeatureExtractor" 25 | PREDICTOR: "FPNPredictor" 26 | ROI_MASK_HEAD: 27 | POOLER_SCALES: (0.25, 0.125, 0.0625, 0.03125) 28 | FEATURE_EXTRACTOR: "MaskRCNNFPNFeatureExtractor" 29 | PREDICTOR: "MaskRCNNC4Predictor" 30 | POOLER_RESOLUTION: 14 31 | POOLER_SAMPLING_RATIO: 2 32 | RESOLUTION: 28 33 | SHARE_BOX_FEATURE_EXTRACTOR: False 34 | MASK_ON: True 35 | DATASETS: 36 | TRAIN: ("coco_2014_train", "coco_2014_valminusminival") 37 | TEST: ("coco_2014_minival",) 38 | DATALOADER: 39 | SIZE_DIVISIBILITY: 32 40 | SOLVER: 41 | BASE_LR: 0.02 42 | WEIGHT_DECAY: 0.0001 43 | STEPS: (120000, 160000) 44 | MAX_ITER: 180000 45 | INPUT: 46 | PIXEL_MEAN: [0.485, 0.456, 0.406] 47 | PIXEL_STD: [0.229, 0.224, 0.225] 48 | TO_BGR255: False 49 | -------------------------------------------------------------------------------- /configs/pytorch_mask_rcnn_R2_50_s2_FPN_2x.yaml: -------------------------------------------------------------------------------- 1 | MODEL: 2 | META_ARCHITECTURE: "GeneralizedRCNN" 3 | WEIGHT: "/home/shgao/.torch/models/res2net50_s2_model_best.pth" 4 | BACKBONE: 5 | CONV_BODY: "R2-50-FPN" 6 | RESNETS: 7 | BACKBONE_OUT_CHANNELS: 256 8 | WIDTH_PER_GROUP: 48 9 | SCALE: 2 10 | TRANS_FUNC: "Bottle2neckWithFixedBatchNorm" 11 | RPN: 12 | USE_FPN: True 13 | ANCHOR_STRIDE: (4, 8, 16, 32, 64) 14 | PRE_NMS_TOP_N_TRAIN: 2000 15 | PRE_NMS_TOP_N_TEST: 1000 16 | POST_NMS_TOP_N_TEST: 1000 17 | FPN_POST_NMS_TOP_N_TEST: 1000 18 | ROI_HEADS: 19 | USE_FPN: True 20 | ROI_BOX_HEAD: 21 | POOLER_RESOLUTION: 7 22 | POOLER_SCALES: (0.25, 0.125, 0.0625, 0.03125) 23 | POOLER_SAMPLING_RATIO: 2 24 | FEATURE_EXTRACTOR: "FPN2MLPFeatureExtractor" 25 | PREDICTOR: "FPNPredictor" 26 | ROI_MASK_HEAD: 27 | POOLER_SCALES: (0.25, 0.125, 0.0625, 0.03125) 28 | FEATURE_EXTRACTOR: "MaskRCNNFPNFeatureExtractor" 29 | PREDICTOR: "MaskRCNNC4Predictor" 30 | POOLER_RESOLUTION: 14 31 | POOLER_SAMPLING_RATIO: 2 32 | RESOLUTION: 28 33 | SHARE_BOX_FEATURE_EXTRACTOR: False 34 | MASK_ON: True 35 | DATASETS: 36 | TRAIN: ("coco_2014_train", "coco_2014_valminusminival") 37 | TEST: ("coco_2014_minival",) 38 | DATALOADER: 39 | SIZE_DIVISIBILITY: 32 40 | SOLVER: 41 | BASE_LR: 0.02 42 | WEIGHT_DECAY: 0.0001 43 | STEPS: (120000, 160000) 44 | MAX_ITER: 180000 45 | INPUT: 46 | PIXEL_MEAN: [0.485, 0.456, 0.406] 47 | PIXEL_STD: [0.229, 0.224, 0.225] 48 | TO_BGR255: False 49 | -------------------------------------------------------------------------------- /configs/pytorch_mask_rcnn_R2_50_s4_FPN_2x.yaml: -------------------------------------------------------------------------------- 1 | MODEL: 2 | META_ARCHITECTURE: "GeneralizedRCNN" 3 | WEIGHT: "/home/shgao/.torch/models/res2net50_26w_4s-06e79181.pth" 4 | BACKBONE: 5 | CONV_BODY: "R2-50-FPN" 6 | RESNETS: 7 | BACKBONE_OUT_CHANNELS: 256 8 | WIDTH_PER_GROUP: 26 9 | SCALE: 4 10 | TRANS_FUNC: "Bottle2neckWithFixedBatchNorm" 11 | RPN: 12 | USE_FPN: True 13 | ANCHOR_STRIDE: (4, 8, 16, 32, 64) 14 | PRE_NMS_TOP_N_TRAIN: 2000 15 | PRE_NMS_TOP_N_TEST: 1000 16 | POST_NMS_TOP_N_TEST: 1000 17 | FPN_POST_NMS_TOP_N_TEST: 1000 18 | ROI_HEADS: 19 | USE_FPN: True 20 | ROI_BOX_HEAD: 21 | POOLER_RESOLUTION: 7 22 | POOLER_SCALES: (0.25, 0.125, 0.0625, 0.03125) 23 | POOLER_SAMPLING_RATIO: 2 24 | FEATURE_EXTRACTOR: "FPN2MLPFeatureExtractor" 25 | PREDICTOR: "FPNPredictor" 26 | ROI_MASK_HEAD: 27 | POOLER_SCALES: (0.25, 0.125, 0.0625, 0.03125) 28 | FEATURE_EXTRACTOR: "MaskRCNNFPNFeatureExtractor" 29 | PREDICTOR: "MaskRCNNC4Predictor" 30 | POOLER_RESOLUTION: 14 31 | POOLER_SAMPLING_RATIO: 2 32 | RESOLUTION: 28 33 | SHARE_BOX_FEATURE_EXTRACTOR: False 34 | MASK_ON: True 35 | DATASETS: 36 | TRAIN: ("coco_2014_train", "coco_2014_valminusminival") 37 | TEST: ("coco_2014_minival",) 38 | DATALOADER: 39 | SIZE_DIVISIBILITY: 32 40 | SOLVER: 41 | BASE_LR: 0.02 42 | WEIGHT_DECAY: 0.0001 43 | STEPS: (120000, 160000) 44 | MAX_ITER: 180000 45 | INPUT: 46 | PIXEL_MEAN: [0.485, 0.456, 0.406] 47 | PIXEL_STD: [0.229, 0.224, 0.225] 48 | TO_BGR255: False 49 | -------------------------------------------------------------------------------- /configs/pytorch_mask_rcnn_R2_50_s6_FPN_2x.yaml: -------------------------------------------------------------------------------- 1 | MODEL: 2 | META_ARCHITECTURE: "GeneralizedRCNN" 3 | WEIGHT: "/home/shgao/.torch/models/res2net50_18w_6s-db17faaa.pth" 4 | BACKBONE: 5 | CONV_BODY: "R2-50-FPN" 6 | RESNETS: 7 | BACKBONE_OUT_CHANNELS: 256 8 | WIDTH_PER_GROUP: 18 9 | SCALE: 6 10 | TRANS_FUNC: "Bottle2neckWithFixedBatchNorm" 11 | RPN: 12 | USE_FPN: True 13 | ANCHOR_STRIDE: (4, 8, 16, 32, 64) 14 | PRE_NMS_TOP_N_TRAIN: 2000 15 | PRE_NMS_TOP_N_TEST: 1000 16 | POST_NMS_TOP_N_TEST: 1000 17 | FPN_POST_NMS_TOP_N_TEST: 1000 18 | ROI_HEADS: 19 | USE_FPN: True 20 | ROI_BOX_HEAD: 21 | POOLER_RESOLUTION: 7 22 | POOLER_SCALES: (0.25, 0.125, 0.0625, 0.03125) 23 | POOLER_SAMPLING_RATIO: 2 24 | FEATURE_EXTRACTOR: "FPN2MLPFeatureExtractor" 25 | PREDICTOR: "FPNPredictor" 26 | ROI_MASK_HEAD: 27 | POOLER_SCALES: (0.25, 0.125, 0.0625, 0.03125) 28 | FEATURE_EXTRACTOR: "MaskRCNNFPNFeatureExtractor" 29 | PREDICTOR: "MaskRCNNC4Predictor" 30 | POOLER_RESOLUTION: 14 31 | POOLER_SAMPLING_RATIO: 2 32 | RESOLUTION: 28 33 | SHARE_BOX_FEATURE_EXTRACTOR: False 34 | MASK_ON: True 35 | DATASETS: 36 | TRAIN: ("coco_2014_train", "coco_2014_valminusminival") 37 | TEST: ("coco_2014_minival",) 38 | DATALOADER: 39 | SIZE_DIVISIBILITY: 32 40 | SOLVER: 41 | BASE_LR: 0.02 42 | WEIGHT_DECAY: 0.0001 43 | STEPS: (120000, 160000) 44 | MAX_ITER: 180000 45 | INPUT: 46 | PIXEL_MEAN: [0.485, 0.456, 0.406] 47 | PIXEL_STD: [0.229, 0.224, 0.225] 48 | TO_BGR255: False 49 | -------------------------------------------------------------------------------- /configs/pytorch_mask_rcnn_R2_50_s8_FPN_2x.yaml: -------------------------------------------------------------------------------- 1 | MODEL: 2 | META_ARCHITECTURE: "GeneralizedRCNN" 3 | WEIGHT: "/home/shgao/.torch/models/res2net50_14w_8s-6527dddc.pth" 4 | BACKBONE: 5 | CONV_BODY: "R2-50-FPN" 6 | RESNETS: 7 | BACKBONE_OUT_CHANNELS: 256 8 | WIDTH_PER_GROUP: 14 9 | SCALE: 8 10 | TRANS_FUNC: "Bottle2neckWithFixedBatchNorm" 11 | RPN: 12 | USE_FPN: True 13 | ANCHOR_STRIDE: (4, 8, 16, 32, 64) 14 | PRE_NMS_TOP_N_TRAIN: 2000 15 | PRE_NMS_TOP_N_TEST: 1000 16 | POST_NMS_TOP_N_TEST: 1000 17 | FPN_POST_NMS_TOP_N_TEST: 1000 18 | ROI_HEADS: 19 | USE_FPN: True 20 | ROI_BOX_HEAD: 21 | POOLER_RESOLUTION: 7 22 | POOLER_SCALES: (0.25, 0.125, 0.0625, 0.03125) 23 | POOLER_SAMPLING_RATIO: 2 24 | FEATURE_EXTRACTOR: "FPN2MLPFeatureExtractor" 25 | PREDICTOR: "FPNPredictor" 26 | ROI_MASK_HEAD: 27 | POOLER_SCALES: (0.25, 0.125, 0.0625, 0.03125) 28 | FEATURE_EXTRACTOR: "MaskRCNNFPNFeatureExtractor" 29 | PREDICTOR: "MaskRCNNC4Predictor" 30 | POOLER_RESOLUTION: 14 31 | POOLER_SAMPLING_RATIO: 2 32 | RESOLUTION: 28 33 | SHARE_BOX_FEATURE_EXTRACTOR: False 34 | MASK_ON: True 35 | DATASETS: 36 | TRAIN: ("coco_2014_train", "coco_2014_valminusminival") 37 | TEST: ("coco_2014_minival",) 38 | DATALOADER: 39 | SIZE_DIVISIBILITY: 32 40 | SOLVER: 41 | BASE_LR: 0.02 42 | WEIGHT_DECAY: 0.0001 43 | STEPS: (120000, 160000) 44 | MAX_ITER: 180000 45 | INPUT: 46 | PIXEL_MEAN: [0.485, 0.456, 0.406] 47 | PIXEL_STD: [0.229, 0.224, 0.225] 48 | TO_BGR255: False 49 | -------------------------------------------------------------------------------- /configs/pytorch_mask_rcnn_R_101_FPN_2x.yaml: -------------------------------------------------------------------------------- 1 | MODEL: 2 | META_ARCHITECTURE: "GeneralizedRCNN" 3 | WEIGHT: "/home/shgao/.torch/models/resnet101-5d3b4d8f.pth" 4 | BACKBONE: 5 | CONV_BODY: "R-101-FPN" 6 | RESNETS: 7 | BACKBONE_OUT_CHANNELS: 256 8 | RPN: 9 | USE_FPN: True 10 | ANCHOR_STRIDE: (4, 8, 16, 32, 64) 11 | PRE_NMS_TOP_N_TRAIN: 2000 12 | PRE_NMS_TOP_N_TEST: 1000 13 | POST_NMS_TOP_N_TEST: 1000 14 | FPN_POST_NMS_TOP_N_TEST: 1000 15 | ROI_HEADS: 16 | USE_FPN: True 17 | ROI_BOX_HEAD: 18 | POOLER_RESOLUTION: 7 19 | POOLER_SCALES: (0.25, 0.125, 0.0625, 0.03125) 20 | POOLER_SAMPLING_RATIO: 2 21 | FEATURE_EXTRACTOR: "FPN2MLPFeatureExtractor" 22 | PREDICTOR: "FPNPredictor" 23 | ROI_MASK_HEAD: 24 | POOLER_SCALES: (0.25, 0.125, 0.0625, 0.03125) 25 | FEATURE_EXTRACTOR: "MaskRCNNFPNFeatureExtractor" 26 | PREDICTOR: "MaskRCNNC4Predictor" 27 | POOLER_RESOLUTION: 14 28 | POOLER_SAMPLING_RATIO: 2 29 | RESOLUTION: 28 30 | SHARE_BOX_FEATURE_EXTRACTOR: False 31 | MASK_ON: True 32 | DATASETS: 33 | TRAIN: ("coco_2014_train", "coco_2014_valminusminival") 34 | TEST: ("coco_2014_minival",) 35 | DATALOADER: 36 | SIZE_DIVISIBILITY: 32 37 | SOLVER: 38 | BASE_LR: 0.02 39 | WEIGHT_DECAY: 0.0001 40 | STEPS: (120000, 160000) 41 | MAX_ITER: 180000 42 | INPUT: 43 | PIXEL_MEAN: [0.485, 0.456, 0.406] 44 | PIXEL_STD: [0.229, 0.224, 0.225] 45 | TO_BGR255: False -------------------------------------------------------------------------------- /configs/pytorch_mask_rcnn_R_50_FPN_2x.yaml: -------------------------------------------------------------------------------- 1 | MODEL: 2 | META_ARCHITECTURE: "GeneralizedRCNN" 3 | WEIGHT: "/home/shgao/.torch/models/resnet50-19c8e357.pth" 4 | BACKBONE: 5 | CONV_BODY: "R-50-FPN" 6 | RESNETS: 7 | BACKBONE_OUT_CHANNELS: 256 8 | RPN: 9 | USE_FPN: True 10 | ANCHOR_STRIDE: (4, 8, 16, 32, 64) 11 | PRE_NMS_TOP_N_TRAIN: 2000 12 | PRE_NMS_TOP_N_TEST: 1000 13 | POST_NMS_TOP_N_TEST: 1000 14 | FPN_POST_NMS_TOP_N_TEST: 1000 15 | ROI_HEADS: 16 | USE_FPN: True 17 | ROI_BOX_HEAD: 18 | POOLER_RESOLUTION: 7 19 | POOLER_SCALES: (0.25, 0.125, 0.0625, 0.03125) 20 | POOLER_SAMPLING_RATIO: 2 21 | FEATURE_EXTRACTOR: "FPN2MLPFeatureExtractor" 22 | PREDICTOR: "FPNPredictor" 23 | ROI_MASK_HEAD: 24 | POOLER_SCALES: (0.25, 0.125, 0.0625, 0.03125) 25 | FEATURE_EXTRACTOR: "MaskRCNNFPNFeatureExtractor" 26 | PREDICTOR: "MaskRCNNC4Predictor" 27 | POOLER_RESOLUTION: 14 28 | POOLER_SAMPLING_RATIO: 2 29 | RESOLUTION: 28 30 | SHARE_BOX_FEATURE_EXTRACTOR: False 31 | MASK_ON: True 32 | DATASETS: 33 | TRAIN: ("coco_2014_train", "coco_2014_valminusminival") 34 | TEST: ("coco_2014_minival",) 35 | DATALOADER: 36 | SIZE_DIVISIBILITY: 32 37 | SOLVER: 38 | BASE_LR: 0.02 39 | WEIGHT_DECAY: 0.0001 40 | STEPS: (120000, 160000) 41 | MAX_ITER: 180000 42 | INPUT: 43 | PIXEL_MEAN: [0.485, 0.456, 0.406] 44 | PIXEL_STD: [0.229, 0.224, 0.225] 45 | TO_BGR255: False -------------------------------------------------------------------------------- /demo/README.md: -------------------------------------------------------------------------------- 1 | ## Webcam and Jupyter notebook demo 2 | 3 | This folder contains a simple webcam demo that illustrates how you can use `maskrcnn_benchmark` for inference. 4 | 5 | 6 | ### With your preferred environment 7 | 8 | You can start it by running it from this folder, using one of the following commands: 9 | ```bash 10 | # by default, it runs on the GPU 11 | # for best results, use min-image-size 800 12 | python webcam.py --min-image-size 800 13 | # can also run it on the CPU 14 | python webcam.py --min-image-size 300 MODEL.DEVICE cpu 15 | # or change the model that you want to use 16 | python webcam.py --config-file ../configs/caffe2/e2e_mask_rcnn_R_101_FPN_1x_caffe2.yaml --min-image-size 300 MODEL.DEVICE cpu 17 | # in order to see the probability heatmaps, pass --show-mask-heatmaps 18 | python webcam.py --min-image-size 300 --show-mask-heatmaps MODEL.DEVICE cpu 19 | ``` 20 | 21 | ### With Docker 22 | 23 | Build the image with the tag `maskrcnn-benchmark` (check [INSTALL.md](../INSTALL.md) for instructions) 24 | 25 | Adjust permissions of the X server host (be careful with this step, refer to 26 | [here](http://wiki.ros.org/docker/Tutorials/GUI) for alternatives) 27 | 28 | ```bash 29 | xhost + 30 | ``` 31 | 32 | Then run a container with the demo: 33 | 34 | ``` 35 | docker run --rm -it \ 36 | -e DISPLAY=${DISPLAY} \ 37 | --privileged \ 38 | -v /tmp/.X11-unix:/tmp/.X11-unix \ 39 | --device=/dev/video0:/dev/video0 \ 40 | --ipc=host maskrcnn-benchmark \ 41 | python demo/webcam.py --min-image-size 300 42 | ``` 43 | 44 | **DISCLAIMER:** *This was tested for an Ubuntu 16.04 machine, 45 | the volume mapping may vary depending on your platform* 46 | -------------------------------------------------------------------------------- /demo/demo_e2e_mask_rcnn_R_50_FPN_1x.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Res2Net/Res2Net-maskrcnn/dd50e0f7b767333335e18fb23054fe27e6b40519/demo/demo_e2e_mask_rcnn_R_50_FPN_1x.png -------------------------------------------------------------------------------- /demo/demo_e2e_mask_rcnn_X_101_32x8d_FPN_1x.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Res2Net/Res2Net-maskrcnn/dd50e0f7b767333335e18fb23054fe27e6b40519/demo/demo_e2e_mask_rcnn_X_101_32x8d_FPN_1x.png -------------------------------------------------------------------------------- /demo/webcam.py: -------------------------------------------------------------------------------- 1 | # Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved. 2 | import argparse 3 | import cv2 4 | 5 | from maskrcnn_benchmark.config import cfg 6 | from predictor import COCODemo 7 | 8 | import time 9 | 10 | 11 | def main(): 12 | parser = argparse.ArgumentParser(description="PyTorch Object Detection Webcam Demo") 13 | parser.add_argument( 14 | "--config-file", 15 | default="../configs/caffe2/e2e_mask_rcnn_R_50_FPN_1x_caffe2.yaml", 16 | metavar="FILE", 17 | help="path to config file", 18 | ) 19 | parser.add_argument( 20 | "--confidence-threshold", 21 | type=float, 22 | default=0.7, 23 | help="Minimum score for the prediction to be shown", 24 | ) 25 | parser.add_argument( 26 | "--min-image-size", 27 | type=int, 28 | default=224, 29 | help="Smallest size of the image to feed to the model. " 30 | "Model was trained with 800, which gives best results", 31 | ) 32 | parser.add_argument( 33 | "--show-mask-heatmaps", 34 | dest="show_mask_heatmaps", 35 | help="Show a heatmap probability for the top masks-per-dim masks", 36 | action="store_true", 37 | ) 38 | parser.add_argument( 39 | "--masks-per-dim", 40 | type=int, 41 | default=2, 42 | help="Number of heatmaps per dimension to show", 43 | ) 44 | parser.add_argument( 45 | "opts", 46 | help="Modify model config options using the command-line", 47 | default=None, 48 | nargs=argparse.REMAINDER, 49 | ) 50 | 51 | args = parser.parse_args() 52 | 53 | # load config from file and command-line arguments 54 | cfg.merge_from_file(args.config_file) 55 | cfg.merge_from_list(args.opts) 56 | cfg.freeze() 57 | 58 | # prepare object that handles inference plus adds predictions on top of image 59 | coco_demo = COCODemo( 60 | cfg, 61 | confidence_threshold=args.confidence_threshold, 62 | show_mask_heatmaps=args.show_mask_heatmaps, 63 | masks_per_dim=args.masks_per_dim, 64 | min_image_size=args.min_image_size, 65 | ) 66 | 67 | cam = cv2.VideoCapture(0) 68 | while True: 69 | start_time = time.time() 70 | ret_val, img = cam.read() 71 | composite = coco_demo.run_on_opencv_image(img) 72 | print("Time: {:.2f} s / img".format(time.time() - start_time)) 73 | cv2.imshow("COCO detections", composite) 74 | if cv2.waitKey(1) == 27: 75 | break # esc to quit 76 | cv2.destroyAllWindows() 77 | 78 | 79 | if __name__ == "__main__": 80 | main() 81 | -------------------------------------------------------------------------------- /docker/Dockerfile: -------------------------------------------------------------------------------- 1 | ARG CUDA="9.0" 2 | ARG CUDNN="7" 3 | 4 | FROM nvidia/cuda:${CUDA}-cudnn${CUDNN}-devel-ubuntu16.04 5 | 6 | RUN echo 'debconf debconf/frontend select Noninteractive' | debconf-set-selections 7 | 8 | # install basics 9 | RUN apt-get update -y \ 10 | && apt-get install -y apt-utils git curl ca-certificates bzip2 cmake tree htop bmon iotop g++ \ 11 | && apt-get install -y libglib2.0-0 libsm6 libxext6 libxrender-dev 12 | 13 | # Install Miniconda 14 | RUN curl -so /miniconda.sh https://repo.continuum.io/miniconda/Miniconda3-latest-Linux-x86_64.sh \ 15 | && chmod +x /miniconda.sh \ 16 | && /miniconda.sh -b -p /miniconda \ 17 | && rm /miniconda.sh 18 | 19 | ENV PATH=/miniconda/bin:$PATH 20 | 21 | # Create a Python 3.6 environment 22 | RUN /miniconda/bin/conda install -y conda-build \ 23 | && /miniconda/bin/conda create -y --name py36 python=3.6.7 \ 24 | && /miniconda/bin/conda clean -ya 25 | 26 | ENV CONDA_DEFAULT_ENV=py36 27 | ENV CONDA_PREFIX=/miniconda/envs/$CONDA_DEFAULT_ENV 28 | ENV PATH=$CONDA_PREFIX/bin:$PATH 29 | ENV CONDA_AUTO_UPDATE_CONDA=false 30 | 31 | RUN conda install -y ipython 32 | RUN pip install ninja yacs cython matplotlib opencv-python 33 | 34 | # Install PyTorch 1.0 Nightly and OpenCV 35 | RUN conda install -y pytorch-nightly -c pytorch \ 36 | && conda clean -ya 37 | 38 | # Install TorchVision master 39 | RUN git clone https://github.com/pytorch/vision.git \ 40 | && cd vision \ 41 | && python setup.py install 42 | 43 | # install pycocotools 44 | RUN git clone https://github.com/cocodataset/cocoapi.git \ 45 | && cd cocoapi/PythonAPI \ 46 | && python setup.py build_ext install 47 | 48 | # install PyTorch Detection 49 | RUN git clone https://github.com/facebookresearch/maskrcnn-benchmark.git \ 50 | && cd maskrcnn-benchmark \ 51 | && python setup.py build develop 52 | 53 | WORKDIR /maskrcnn-benchmark 54 | -------------------------------------------------------------------------------- /docker/docker-jupyter/Dockerfile: -------------------------------------------------------------------------------- 1 | ARG CUDA="9.0" 2 | ARG CUDNN="7" 3 | 4 | FROM nvidia/cuda:${CUDA}-cudnn${CUDNN}-devel-ubuntu16.04 5 | 6 | RUN echo 'debconf debconf/frontend select Noninteractive' | debconf-set-selections 7 | 8 | # install basics 9 | RUN apt-get update -y \ 10 | && apt-get install -y apt-utils git curl ca-certificates bzip2 cmake tree htop bmon iotop g++ 11 | 12 | # Install Miniconda 13 | RUN curl -so /miniconda.sh https://repo.continuum.io/miniconda/Miniconda3-latest-Linux-x86_64.sh \ 14 | && chmod +x /miniconda.sh \ 15 | && /miniconda.sh -b -p /miniconda \ 16 | && rm /miniconda.sh 17 | 18 | ENV PATH=/miniconda/bin:$PATH 19 | 20 | # Create a Python 3.6 environment 21 | RUN /miniconda/bin/conda install -y conda-build \ 22 | && /miniconda/bin/conda create -y --name py36 python=3.6.7 \ 23 | && /miniconda/bin/conda clean -ya 24 | 25 | ENV CONDA_DEFAULT_ENV=py36 26 | ENV CONDA_PREFIX=/miniconda/envs/$CONDA_DEFAULT_ENV 27 | ENV PATH=$CONDA_PREFIX/bin:$PATH 28 | ENV CONDA_AUTO_UPDATE_CONDA=false 29 | 30 | RUN conda install -y ipython 31 | RUN pip install ninja yacs cython matplotlib jupyter 32 | 33 | # Install PyTorch 1.0 Nightly and OpenCV 34 | RUN conda install -y pytorch-nightly -c pytorch \ 35 | && conda install -y opencv -c menpo \ 36 | && conda clean -ya 37 | 38 | WORKDIR /root 39 | 40 | USER root 41 | 42 | RUN mkdir /notebooks 43 | 44 | WORKDIR /notebooks 45 | 46 | # Install TorchVision master 47 | RUN git clone https://github.com/pytorch/vision.git \ 48 | && cd vision \ 49 | && python setup.py install 50 | 51 | # install pycocotools 52 | RUN git clone https://github.com/cocodataset/cocoapi.git \ 53 | && cd cocoapi/PythonAPI \ 54 | && python setup.py build_ext install 55 | 56 | # install PyTorch Detection 57 | RUN git clone https://github.com/facebookresearch/maskrcnn-benchmark.git \ 58 | && cd maskrcnn-benchmark \ 59 | && python setup.py build develop 60 | 61 | RUN jupyter notebook --generate-config 62 | 63 | ENV CONFIG_PATH="/root/.jupyter/jupyter_notebook_config.py" 64 | 65 | COPY "jupyter_notebook_config.py" ${CONFIG_PATH} 66 | 67 | ENTRYPOINT ["sh", "-c", "jupyter notebook --allow-root -y --no-browser --ip=0.0.0.0 --config=${CONFIG_PATH}"] 68 | -------------------------------------------------------------------------------- /docker/docker-jupyter/jupyter_notebook_config.py: -------------------------------------------------------------------------------- 1 | import os 2 | from IPython.lib import passwd 3 | 4 | #c = c # pylint:disable=undefined-variable 5 | c = get_config() 6 | c.NotebookApp.ip = '0.0.0.0' 7 | c.NotebookApp.port = int(os.getenv('PORT', 8888)) 8 | c.NotebookApp.open_browser = False 9 | 10 | # sets a password if PASSWORD is set in the environment 11 | if 'PASSWORD' in os.environ: 12 | password = os.environ['PASSWORD'] 13 | if password: 14 | c.NotebookApp.password = passwd(password) 15 | else: 16 | c.NotebookApp.password = '' 17 | c.NotebookApp.token = '' 18 | del os.environ['PASSWORD'] 19 | -------------------------------------------------------------------------------- /maskrcnn_benchmark/__init__.py: -------------------------------------------------------------------------------- 1 | # Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved. 2 | -------------------------------------------------------------------------------- /maskrcnn_benchmark/config/__init__.py: -------------------------------------------------------------------------------- 1 | # Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved. 2 | from .defaults import _C as cfg 3 | -------------------------------------------------------------------------------- /maskrcnn_benchmark/csrc/ROIAlign.h: -------------------------------------------------------------------------------- 1 | // Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved. 2 | #pragma once 3 | 4 | #include "cpu/vision.h" 5 | 6 | #ifdef WITH_CUDA 7 | #include "cuda/vision.h" 8 | #endif 9 | 10 | // Interface for Python 11 | at::Tensor ROIAlign_forward(const at::Tensor& input, 12 | const at::Tensor& rois, 13 | const float spatial_scale, 14 | const int pooled_height, 15 | const int pooled_width, 16 | const int sampling_ratio) { 17 | if (input.type().is_cuda()) { 18 | #ifdef WITH_CUDA 19 | return ROIAlign_forward_cuda(input, rois, spatial_scale, pooled_height, pooled_width, sampling_ratio); 20 | #else 21 | AT_ERROR("Not compiled with GPU support"); 22 | #endif 23 | } 24 | return ROIAlign_forward_cpu(input, rois, spatial_scale, pooled_height, pooled_width, sampling_ratio); 25 | } 26 | 27 | at::Tensor ROIAlign_backward(const at::Tensor& grad, 28 | const at::Tensor& rois, 29 | const float spatial_scale, 30 | const int pooled_height, 31 | const int pooled_width, 32 | const int batch_size, 33 | const int channels, 34 | const int height, 35 | const int width, 36 | const int sampling_ratio) { 37 | if (grad.type().is_cuda()) { 38 | #ifdef WITH_CUDA 39 | return ROIAlign_backward_cuda(grad, rois, spatial_scale, pooled_height, pooled_width, batch_size, channels, height, width, sampling_ratio); 40 | #else 41 | AT_ERROR("Not compiled with GPU support"); 42 | #endif 43 | } 44 | AT_ERROR("Not implemented on the CPU"); 45 | } 46 | 47 | -------------------------------------------------------------------------------- /maskrcnn_benchmark/csrc/ROIPool.h: -------------------------------------------------------------------------------- 1 | // Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved. 2 | #pragma once 3 | 4 | #include "cpu/vision.h" 5 | 6 | #ifdef WITH_CUDA 7 | #include "cuda/vision.h" 8 | #endif 9 | 10 | 11 | std::tuple ROIPool_forward(const at::Tensor& input, 12 | const at::Tensor& rois, 13 | const float spatial_scale, 14 | const int pooled_height, 15 | const int pooled_width) { 16 | if (input.type().is_cuda()) { 17 | #ifdef WITH_CUDA 18 | return ROIPool_forward_cuda(input, rois, spatial_scale, pooled_height, pooled_width); 19 | #else 20 | AT_ERROR("Not compiled with GPU support"); 21 | #endif 22 | } 23 | AT_ERROR("Not implemented on the CPU"); 24 | } 25 | 26 | at::Tensor ROIPool_backward(const at::Tensor& grad, 27 | const at::Tensor& input, 28 | const at::Tensor& rois, 29 | const at::Tensor& argmax, 30 | const float spatial_scale, 31 | const int pooled_height, 32 | const int pooled_width, 33 | const int batch_size, 34 | const int channels, 35 | const int height, 36 | const int width) { 37 | if (grad.type().is_cuda()) { 38 | #ifdef WITH_CUDA 39 | return ROIPool_backward_cuda(grad, input, rois, argmax, spatial_scale, pooled_height, pooled_width, batch_size, channels, height, width); 40 | #else 41 | AT_ERROR("Not compiled with GPU support"); 42 | #endif 43 | } 44 | AT_ERROR("Not implemented on the CPU"); 45 | } 46 | 47 | 48 | 49 | -------------------------------------------------------------------------------- /maskrcnn_benchmark/csrc/SigmoidFocalLoss.h: -------------------------------------------------------------------------------- 1 | #pragma once 2 | 3 | #include "cpu/vision.h" 4 | 5 | #ifdef WITH_CUDA 6 | #include "cuda/vision.h" 7 | #endif 8 | 9 | // Interface for Python 10 | at::Tensor SigmoidFocalLoss_forward( 11 | const at::Tensor& logits, 12 | const at::Tensor& targets, 13 | const int num_classes, 14 | const float gamma, 15 | const float alpha) { 16 | if (logits.type().is_cuda()) { 17 | #ifdef WITH_CUDA 18 | return SigmoidFocalLoss_forward_cuda(logits, targets, num_classes, gamma, alpha); 19 | #else 20 | AT_ERROR("Not compiled with GPU support"); 21 | #endif 22 | } 23 | AT_ERROR("Not implemented on the CPU"); 24 | } 25 | 26 | at::Tensor SigmoidFocalLoss_backward( 27 | const at::Tensor& logits, 28 | const at::Tensor& targets, 29 | const at::Tensor& d_losses, 30 | const int num_classes, 31 | const float gamma, 32 | const float alpha) { 33 | if (logits.type().is_cuda()) { 34 | #ifdef WITH_CUDA 35 | return SigmoidFocalLoss_backward_cuda(logits, targets, d_losses, num_classes, gamma, alpha); 36 | #else 37 | AT_ERROR("Not compiled with GPU support"); 38 | #endif 39 | } 40 | AT_ERROR("Not implemented on the CPU"); 41 | } 42 | -------------------------------------------------------------------------------- /maskrcnn_benchmark/csrc/cpu/nms_cpu.cpp: -------------------------------------------------------------------------------- 1 | // Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved. 2 | #include "cpu/vision.h" 3 | 4 | 5 | template 6 | at::Tensor nms_cpu_kernel(const at::Tensor& dets, 7 | const at::Tensor& scores, 8 | const float threshold) { 9 | AT_ASSERTM(!dets.type().is_cuda(), "dets must be a CPU tensor"); 10 | AT_ASSERTM(!scores.type().is_cuda(), "scores must be a CPU tensor"); 11 | AT_ASSERTM(dets.type() == scores.type(), "dets should have the same type as scores"); 12 | 13 | if (dets.numel() == 0) { 14 | return at::empty({0}, dets.options().dtype(at::kLong).device(at::kCPU)); 15 | } 16 | 17 | auto x1_t = dets.select(1, 0).contiguous(); 18 | auto y1_t = dets.select(1, 1).contiguous(); 19 | auto x2_t = dets.select(1, 2).contiguous(); 20 | auto y2_t = dets.select(1, 3).contiguous(); 21 | 22 | at::Tensor areas_t = (x2_t - x1_t + 1) * (y2_t - y1_t + 1); 23 | 24 | auto order_t = std::get<1>(scores.sort(0, /* descending=*/true)); 25 | 26 | auto ndets = dets.size(0); 27 | at::Tensor suppressed_t = at::zeros({ndets}, dets.options().dtype(at::kByte).device(at::kCPU)); 28 | 29 | auto suppressed = suppressed_t.data(); 30 | auto order = order_t.data(); 31 | auto x1 = x1_t.data(); 32 | auto y1 = y1_t.data(); 33 | auto x2 = x2_t.data(); 34 | auto y2 = y2_t.data(); 35 | auto areas = areas_t.data(); 36 | 37 | for (int64_t _i = 0; _i < ndets; _i++) { 38 | auto i = order[_i]; 39 | if (suppressed[i] == 1) 40 | continue; 41 | auto ix1 = x1[i]; 42 | auto iy1 = y1[i]; 43 | auto ix2 = x2[i]; 44 | auto iy2 = y2[i]; 45 | auto iarea = areas[i]; 46 | 47 | for (int64_t _j = _i + 1; _j < ndets; _j++) { 48 | auto j = order[_j]; 49 | if (suppressed[j] == 1) 50 | continue; 51 | auto xx1 = std::max(ix1, x1[j]); 52 | auto yy1 = std::max(iy1, y1[j]); 53 | auto xx2 = std::min(ix2, x2[j]); 54 | auto yy2 = std::min(iy2, y2[j]); 55 | 56 | auto w = std::max(static_cast(0), xx2 - xx1 + 1); 57 | auto h = std::max(static_cast(0), yy2 - yy1 + 1); 58 | auto inter = w * h; 59 | auto ovr = inter / (iarea + areas[j] - inter); 60 | if (ovr >= threshold) 61 | suppressed[j] = 1; 62 | } 63 | } 64 | return at::nonzero(suppressed_t == 0).squeeze(1); 65 | } 66 | 67 | at::Tensor nms_cpu(const at::Tensor& dets, 68 | const at::Tensor& scores, 69 | const float threshold) { 70 | at::Tensor result; 71 | AT_DISPATCH_FLOATING_TYPES(dets.type(), "nms", [&] { 72 | result = nms_cpu_kernel(dets, scores, threshold); 73 | }); 74 | return result; 75 | } 76 | -------------------------------------------------------------------------------- /maskrcnn_benchmark/csrc/cpu/vision.h: -------------------------------------------------------------------------------- 1 | // Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved. 2 | #pragma once 3 | #include 4 | 5 | 6 | at::Tensor ROIAlign_forward_cpu(const at::Tensor& input, 7 | const at::Tensor& rois, 8 | const float spatial_scale, 9 | const int pooled_height, 10 | const int pooled_width, 11 | const int sampling_ratio); 12 | 13 | 14 | at::Tensor nms_cpu(const at::Tensor& dets, 15 | const at::Tensor& scores, 16 | const float threshold); 17 | -------------------------------------------------------------------------------- /maskrcnn_benchmark/csrc/cuda/vision.h: -------------------------------------------------------------------------------- 1 | // Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved. 2 | #pragma once 3 | #include 4 | 5 | 6 | at::Tensor SigmoidFocalLoss_forward_cuda( 7 | const at::Tensor& logits, 8 | const at::Tensor& targets, 9 | const int num_classes, 10 | const float gamma, 11 | const float alpha); 12 | 13 | at::Tensor SigmoidFocalLoss_backward_cuda( 14 | const at::Tensor& logits, 15 | const at::Tensor& targets, 16 | const at::Tensor& d_losses, 17 | const int num_classes, 18 | const float gamma, 19 | const float alpha); 20 | 21 | at::Tensor ROIAlign_forward_cuda(const at::Tensor& input, 22 | const at::Tensor& rois, 23 | const float spatial_scale, 24 | const int pooled_height, 25 | const int pooled_width, 26 | const int sampling_ratio); 27 | 28 | at::Tensor ROIAlign_backward_cuda(const at::Tensor& grad, 29 | const at::Tensor& rois, 30 | const float spatial_scale, 31 | const int pooled_height, 32 | const int pooled_width, 33 | const int batch_size, 34 | const int channels, 35 | const int height, 36 | const int width, 37 | const int sampling_ratio); 38 | 39 | 40 | std::tuple ROIPool_forward_cuda(const at::Tensor& input, 41 | const at::Tensor& rois, 42 | const float spatial_scale, 43 | const int pooled_height, 44 | const int pooled_width); 45 | 46 | at::Tensor ROIPool_backward_cuda(const at::Tensor& grad, 47 | const at::Tensor& input, 48 | const at::Tensor& rois, 49 | const at::Tensor& argmax, 50 | const float spatial_scale, 51 | const int pooled_height, 52 | const int pooled_width, 53 | const int batch_size, 54 | const int channels, 55 | const int height, 56 | const int width); 57 | 58 | at::Tensor nms_cuda(const at::Tensor boxes, float nms_overlap_thresh); 59 | 60 | 61 | at::Tensor compute_flow_cuda(const at::Tensor& boxes, 62 | const int height, 63 | const int width); 64 | -------------------------------------------------------------------------------- /maskrcnn_benchmark/csrc/nms.h: -------------------------------------------------------------------------------- 1 | // Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved. 2 | #pragma once 3 | #include "cpu/vision.h" 4 | 5 | #ifdef WITH_CUDA 6 | #include "cuda/vision.h" 7 | #endif 8 | 9 | 10 | at::Tensor nms(const at::Tensor& dets, 11 | const at::Tensor& scores, 12 | const float threshold) { 13 | 14 | if (dets.type().is_cuda()) { 15 | #ifdef WITH_CUDA 16 | // TODO raise error if not compiled with CUDA 17 | if (dets.numel() == 0) 18 | return at::empty({0}, dets.options().dtype(at::kLong).device(at::kCPU)); 19 | auto b = at::cat({dets, scores.unsqueeze(1)}, 1); 20 | return nms_cuda(b, threshold); 21 | #else 22 | AT_ERROR("Not compiled with GPU support"); 23 | #endif 24 | } 25 | 26 | at::Tensor result = nms_cpu(dets, scores, threshold); 27 | return result; 28 | } 29 | -------------------------------------------------------------------------------- /maskrcnn_benchmark/csrc/vision.cpp: -------------------------------------------------------------------------------- 1 | // Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved. 2 | #include "nms.h" 3 | #include "ROIAlign.h" 4 | #include "ROIPool.h" 5 | #include "SigmoidFocalLoss.h" 6 | 7 | PYBIND11_MODULE(TORCH_EXTENSION_NAME, m) { 8 | m.def("nms", &nms, "non-maximum suppression"); 9 | m.def("roi_align_forward", &ROIAlign_forward, "ROIAlign_forward"); 10 | m.def("roi_align_backward", &ROIAlign_backward, "ROIAlign_backward"); 11 | m.def("roi_pool_forward", &ROIPool_forward, "ROIPool_forward"); 12 | m.def("roi_pool_backward", &ROIPool_backward, "ROIPool_backward"); 13 | m.def("sigmoid_focalloss_forward", &SigmoidFocalLoss_forward, "SigmoidFocalLoss_forward"); 14 | m.def("sigmoid_focalloss_backward", &SigmoidFocalLoss_backward, "SigmoidFocalLoss_backward"); 15 | } 16 | -------------------------------------------------------------------------------- /maskrcnn_benchmark/data/README.md: -------------------------------------------------------------------------------- 1 | # Setting Up Datasets 2 | This file describes how to perform training on other datasets. 3 | 4 | Only Pascal VOC dataset can be loaded from its original format and be outputted to Pascal style results currently. 5 | 6 | We expect the annotations from other datasets be converted to COCO json format, and 7 | the output will be in COCO-style. (i.e. AP, AP50, AP75, APs, APm, APl for bbox and segm) 8 | 9 | ## Creating Symlinks for PASCAL VOC 10 | 11 | We assume that your symlinked `datasets/voc/VOC` directory has the following structure: 12 | 13 | ``` 14 | VOC 15 | |_ JPEGImages 16 | | |_ .jpg 17 | | |_ ... 18 | | |_ .jpg 19 | |_ Annotations 20 | | |_ pascal_train.json (optional) 21 | | |_ pascal_val.json (optional) 22 | | |_ pascal_test.json (optional) 23 | | |_ .xml 24 | | |_ ... 25 | | |_ .xml 26 | |_ VOCdevkit 27 | ``` 28 | 29 | Create symlinks for `voc/VOC`: 30 | 31 | ``` 32 | cd ~/github/maskrcnn-benchmark 33 | mkdir -p datasets/voc/VOC 34 | ln -s /path/to/VOC /datasets/voc/VOC 35 | ``` 36 | Example configuration files for PASCAL VOC could be found [here](https://github.com/facebookresearch/maskrcnn-benchmark/blob/master/configs/pascal_voc/). 37 | 38 | ### PASCAL VOC Annotations in COCO Format 39 | To output COCO-style evaluation result, PASCAL VOC annotations in COCO json format is required and could be downloaded from [here](https://storage.googleapis.com/coco-dataset/external/PASCAL_VOC.zip) 40 | via http://cocodataset.org/#external. 41 | 42 | ## Creating Symlinks for Cityscapes: 43 | 44 | We assume that your symlinked `datasets/cityscapes` directory has the following structure: 45 | 46 | ``` 47 | cityscapes 48 | |_ images 49 | | |_ .jpg 50 | | |_ ... 51 | | |_ .jpg 52 | |_ annotations 53 | | |_ instanceonly_gtFile_train.json 54 | | |_ ... 55 | |_ raw 56 | |_ gtFine 57 | |_ ... 58 | |_ README.md 59 | ``` 60 | 61 | Create symlinks for `cityscapes`: 62 | 63 | ``` 64 | cd ~/github/maskrcnn-benchmark 65 | mkdir -p datasets/cityscapes 66 | ln -s /path/to/cityscapes datasets/data/cityscapes 67 | ``` 68 | 69 | ### Steps to convert Cityscapes Annotations to COCO Format 70 | 1. Download gtFine_trainvaltest.zip from https://www.cityscapes-dataset.com/downloads/ (login required) 71 | 2. Extract it to /path/to/gtFine_trainvaltest 72 | ``` 73 | cityscapes 74 | |_ gtFine_trainvaltest.zip 75 | |_ gtFine_trainvaltest 76 | |_ gtFine 77 | ``` 78 | 3. Run the below commands to convert the annotations 79 | 80 | ``` 81 | cd ~/github 82 | git clone https://github.com/mcordts/cityscapesScripts.git 83 | cd cityscapesScripts 84 | cp ~/github/maskrcnn-benchmark/tools/cityscapes/instances2dict_with_polygons.py cityscapesscripts/evaluation 85 | python setup.py install 86 | cd ~/github/maskrcnn-benchmark 87 | python tools/cityscapes/convert_cityscapes_to_coco.py --datadir /path/to/cityscapes --outdir /path/to/cityscapes/annotations 88 | ``` 89 | 90 | Example configuration files for Cityscapes could be found [here](https://github.com/facebookresearch/maskrcnn-benchmark/blob/master/configs/cityscapes/). 91 | -------------------------------------------------------------------------------- /maskrcnn_benchmark/data/__init__.py: -------------------------------------------------------------------------------- 1 | # Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved. 2 | from .build import make_data_loader 3 | -------------------------------------------------------------------------------- /maskrcnn_benchmark/data/collate_batch.py: -------------------------------------------------------------------------------- 1 | # Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved. 2 | from maskrcnn_benchmark.structures.image_list import to_image_list 3 | 4 | 5 | class BatchCollator(object): 6 | """ 7 | From a list of samples from the dataset, 8 | returns the batched images and targets. 9 | This should be passed to the DataLoader 10 | """ 11 | 12 | def __init__(self, size_divisible=0): 13 | self.size_divisible = size_divisible 14 | 15 | def __call__(self, batch): 16 | transposed_batch = list(zip(*batch)) 17 | images = to_image_list(transposed_batch[0], self.size_divisible) 18 | targets = transposed_batch[1] 19 | img_ids = transposed_batch[2] 20 | return images, targets, img_ids 21 | -------------------------------------------------------------------------------- /maskrcnn_benchmark/data/datasets/__init__.py: -------------------------------------------------------------------------------- 1 | # Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved. 2 | from .coco import COCODataset 3 | from .voc import PascalVOCDataset 4 | from .concat_dataset import ConcatDataset 5 | 6 | __all__ = ["COCODataset", "ConcatDataset", "PascalVOCDataset"] 7 | -------------------------------------------------------------------------------- /maskrcnn_benchmark/data/datasets/coco.py: -------------------------------------------------------------------------------- 1 | # Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved. 2 | import torch 3 | import torchvision 4 | 5 | from maskrcnn_benchmark.structures.bounding_box import BoxList 6 | from maskrcnn_benchmark.structures.segmentation_mask import SegmentationMask 7 | from maskrcnn_benchmark.structures.keypoint import PersonKeypoints 8 | 9 | 10 | min_keypoints_per_image = 10 11 | 12 | 13 | def _count_visible_keypoints(anno): 14 | return sum(sum(1 for v in ann["keypoints"][2::3] if v > 0) for ann in anno) 15 | 16 | 17 | def _has_only_empty_bbox(anno): 18 | return all(any(o <= 1 for o in obj["bbox"][2:]) for obj in anno) 19 | 20 | 21 | def has_valid_annotation(anno): 22 | # if it's empty, there is no annotation 23 | if len(anno) == 0: 24 | return False 25 | # if all boxes have close to zero area, there is no annotation 26 | if _has_only_empty_bbox(anno): 27 | return False 28 | # keypoints task have a slight different critera for considering 29 | # if an annotation is valid 30 | if "keypoints" not in anno[0]: 31 | return True 32 | # for keypoint detection tasks, only consider valid images those 33 | # containing at least min_keypoints_per_image 34 | if _count_visible_keypoints(anno) >= min_keypoints_per_image: 35 | return True 36 | return False 37 | 38 | 39 | class COCODataset(torchvision.datasets.coco.CocoDetection): 40 | def __init__( 41 | self, ann_file, root, remove_images_without_annotations, transforms=None 42 | ): 43 | super(COCODataset, self).__init__(root, ann_file) 44 | # sort indices for reproducible results 45 | self.ids = sorted(self.ids) 46 | 47 | # filter images without detection annotations 48 | if remove_images_without_annotations: 49 | ids = [] 50 | for img_id in self.ids: 51 | ann_ids = self.coco.getAnnIds(imgIds=img_id, iscrowd=None) 52 | anno = self.coco.loadAnns(ann_ids) 53 | if has_valid_annotation(anno): 54 | ids.append(img_id) 55 | self.ids = ids 56 | 57 | self.json_category_id_to_contiguous_id = { 58 | v: i + 1 for i, v in enumerate(self.coco.getCatIds()) 59 | } 60 | self.contiguous_category_id_to_json_id = { 61 | v: k for k, v in self.json_category_id_to_contiguous_id.items() 62 | } 63 | self.id_to_img_map = {k: v for k, v in enumerate(self.ids)} 64 | self.transforms = transforms 65 | 66 | def __getitem__(self, idx): 67 | img, anno = super(COCODataset, self).__getitem__(idx) 68 | 69 | # filter crowd annotations 70 | # TODO might be better to add an extra field 71 | anno = [obj for obj in anno if obj["iscrowd"] == 0] 72 | 73 | boxes = [obj["bbox"] for obj in anno] 74 | boxes = torch.as_tensor(boxes).reshape(-1, 4) # guard against no boxes 75 | target = BoxList(boxes, img.size, mode="xywh").convert("xyxy") 76 | 77 | classes = [obj["category_id"] for obj in anno] 78 | classes = [self.json_category_id_to_contiguous_id[c] for c in classes] 79 | classes = torch.tensor(classes) 80 | target.add_field("labels", classes) 81 | 82 | masks = [obj["segmentation"] for obj in anno] 83 | masks = SegmentationMask(masks, img.size) 84 | target.add_field("masks", masks) 85 | 86 | if anno and "keypoints" in anno[0]: 87 | keypoints = [obj["keypoints"] for obj in anno] 88 | keypoints = PersonKeypoints(keypoints, img.size) 89 | target.add_field("keypoints", keypoints) 90 | 91 | target = target.clip_to_image(remove_empty=True) 92 | 93 | if self.transforms is not None: 94 | img, target = self.transforms(img, target) 95 | 96 | return img, target, idx 97 | 98 | def get_img_info(self, index): 99 | img_id = self.id_to_img_map[index] 100 | img_data = self.coco.imgs[img_id] 101 | return img_data 102 | -------------------------------------------------------------------------------- /maskrcnn_benchmark/data/datasets/concat_dataset.py: -------------------------------------------------------------------------------- 1 | # Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved. 2 | import bisect 3 | 4 | from torch.utils.data.dataset import ConcatDataset as _ConcatDataset 5 | 6 | 7 | class ConcatDataset(_ConcatDataset): 8 | """ 9 | Same as torch.utils.data.dataset.ConcatDataset, but exposes an extra 10 | method for querying the sizes of the image 11 | """ 12 | 13 | def get_idxs(self, idx): 14 | dataset_idx = bisect.bisect_right(self.cumulative_sizes, idx) 15 | if dataset_idx == 0: 16 | sample_idx = idx 17 | else: 18 | sample_idx = idx - self.cumulative_sizes[dataset_idx - 1] 19 | return dataset_idx, sample_idx 20 | 21 | def get_img_info(self, idx): 22 | dataset_idx, sample_idx = self.get_idxs(idx) 23 | return self.datasets[dataset_idx].get_img_info(sample_idx) 24 | -------------------------------------------------------------------------------- /maskrcnn_benchmark/data/datasets/evaluation/__init__.py: -------------------------------------------------------------------------------- 1 | from maskrcnn_benchmark.data import datasets 2 | 3 | from .coco import coco_evaluation 4 | from .voc import voc_evaluation 5 | 6 | 7 | def evaluate(dataset, predictions, output_folder, **kwargs): 8 | """evaluate dataset using different methods based on dataset type. 9 | Args: 10 | dataset: Dataset object 11 | predictions(list[BoxList]): each item in the list represents the 12 | prediction results for one image. 13 | output_folder: output folder, to save evaluation files or results. 14 | **kwargs: other args. 15 | Returns: 16 | evaluation result 17 | """ 18 | args = dict( 19 | dataset=dataset, predictions=predictions, output_folder=output_folder, **kwargs 20 | ) 21 | if isinstance(dataset, datasets.COCODataset): 22 | return coco_evaluation(**args) 23 | elif isinstance(dataset, datasets.PascalVOCDataset): 24 | return voc_evaluation(**args) 25 | else: 26 | dataset_name = dataset.__class__.__name__ 27 | raise NotImplementedError("Unsupported dataset type {}.".format(dataset_name)) 28 | -------------------------------------------------------------------------------- /maskrcnn_benchmark/data/datasets/evaluation/coco/__init__.py: -------------------------------------------------------------------------------- 1 | from .coco_eval import do_coco_evaluation 2 | 3 | 4 | def coco_evaluation( 5 | dataset, 6 | predictions, 7 | output_folder, 8 | box_only, 9 | iou_types, 10 | expected_results, 11 | expected_results_sigma_tol, 12 | ): 13 | return do_coco_evaluation( 14 | dataset=dataset, 15 | predictions=predictions, 16 | box_only=box_only, 17 | output_folder=output_folder, 18 | iou_types=iou_types, 19 | expected_results=expected_results, 20 | expected_results_sigma_tol=expected_results_sigma_tol, 21 | ) 22 | -------------------------------------------------------------------------------- /maskrcnn_benchmark/data/datasets/evaluation/voc/__init__.py: -------------------------------------------------------------------------------- 1 | import logging 2 | 3 | from .voc_eval import do_voc_evaluation 4 | 5 | 6 | def voc_evaluation(dataset, predictions, output_folder, box_only, **_): 7 | logger = logging.getLogger("maskrcnn_benchmark.inference") 8 | if box_only: 9 | logger.warning("voc evaluation doesn't support box_only, ignored.") 10 | logger.info("performing voc evaluation, ignored iou_types.") 11 | return do_voc_evaluation( 12 | dataset=dataset, 13 | predictions=predictions, 14 | output_folder=output_folder, 15 | logger=logger, 16 | ) 17 | -------------------------------------------------------------------------------- /maskrcnn_benchmark/data/datasets/list_dataset.py: -------------------------------------------------------------------------------- 1 | # Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved. 2 | """ 3 | Simple dataset class that wraps a list of path names 4 | """ 5 | 6 | from PIL import Image 7 | 8 | from maskrcnn_benchmark.structures.bounding_box import BoxList 9 | 10 | 11 | class ListDataset(object): 12 | def __init__(self, image_lists, transforms=None): 13 | self.image_lists = image_lists 14 | self.transforms = transforms 15 | 16 | def __getitem__(self, item): 17 | img = Image.open(self.image_lists[item]).convert("RGB") 18 | 19 | # dummy target 20 | w, h = img.size 21 | target = BoxList([[0, 0, w, h]], img.size, mode="xyxy") 22 | 23 | if self.transforms is not None: 24 | img, target = self.transforms(img, target) 25 | 26 | return img, target 27 | 28 | def __len__(self): 29 | return len(self.image_lists) 30 | 31 | def get_img_info(self, item): 32 | """ 33 | Return the image dimensions for the image, without 34 | loading and pre-processing it 35 | """ 36 | pass 37 | -------------------------------------------------------------------------------- /maskrcnn_benchmark/data/datasets/voc.py: -------------------------------------------------------------------------------- 1 | import os 2 | 3 | import torch 4 | import torch.utils.data 5 | from PIL import Image 6 | import sys 7 | 8 | if sys.version_info[0] == 2: 9 | import xml.etree.cElementTree as ET 10 | else: 11 | import xml.etree.ElementTree as ET 12 | 13 | 14 | from maskrcnn_benchmark.structures.bounding_box import BoxList 15 | 16 | 17 | class PascalVOCDataset(torch.utils.data.Dataset): 18 | 19 | CLASSES = ( 20 | "__background__ ", 21 | "aeroplane", 22 | "bicycle", 23 | "bird", 24 | "boat", 25 | "bottle", 26 | "bus", 27 | "car", 28 | "cat", 29 | "chair", 30 | "cow", 31 | "diningtable", 32 | "dog", 33 | "horse", 34 | "motorbike", 35 | "person", 36 | "pottedplant", 37 | "sheep", 38 | "sofa", 39 | "train", 40 | "tvmonitor", 41 | ) 42 | 43 | def __init__(self, data_dir, split, use_difficult=False, transforms=None): 44 | self.root = data_dir 45 | self.image_set = split 46 | self.keep_difficult = use_difficult 47 | self.transforms = transforms 48 | 49 | self._annopath = os.path.join(self.root, "Annotations", "%s.xml") 50 | self._imgpath = os.path.join(self.root, "JPEGImages", "%s.jpg") 51 | self._imgsetpath = os.path.join(self.root, "ImageSets", "Main", "%s.txt") 52 | 53 | with open(self._imgsetpath % self.image_set) as f: 54 | self.ids = f.readlines() 55 | self.ids = [x.strip("\n") for x in self.ids] 56 | self.id_to_img_map = {k: v for k, v in enumerate(self.ids)} 57 | 58 | cls = PascalVOCDataset.CLASSES 59 | self.class_to_ind = dict(zip(cls, range(len(cls)))) 60 | 61 | def __getitem__(self, index): 62 | img_id = self.ids[index] 63 | img = Image.open(self._imgpath % img_id).convert("RGB") 64 | 65 | target = self.get_groundtruth(index) 66 | target = target.clip_to_image(remove_empty=True) 67 | 68 | if self.transforms is not None: 69 | img, target = self.transforms(img, target) 70 | 71 | return img, target, index 72 | 73 | def __len__(self): 74 | return len(self.ids) 75 | 76 | def get_groundtruth(self, index): 77 | img_id = self.ids[index] 78 | anno = ET.parse(self._annopath % img_id).getroot() 79 | anno = self._preprocess_annotation(anno) 80 | 81 | height, width = anno["im_info"] 82 | target = BoxList(anno["boxes"], (width, height), mode="xyxy") 83 | target.add_field("labels", anno["labels"]) 84 | target.add_field("difficult", anno["difficult"]) 85 | return target 86 | 87 | def _preprocess_annotation(self, target): 88 | boxes = [] 89 | gt_classes = [] 90 | difficult_boxes = [] 91 | TO_REMOVE = 1 92 | 93 | for obj in target.iter("object"): 94 | difficult = int(obj.find("difficult").text) == 1 95 | if not self.keep_difficult and difficult: 96 | continue 97 | name = obj.find("name").text.lower().strip() 98 | bb = obj.find("bndbox") 99 | # Make pixel indexes 0-based 100 | # Refer to "https://github.com/rbgirshick/py-faster-rcnn/blob/master/lib/datasets/pascal_voc.py#L208-L211" 101 | box = [ 102 | bb.find("xmin").text, 103 | bb.find("ymin").text, 104 | bb.find("xmax").text, 105 | bb.find("ymax").text, 106 | ] 107 | bndbox = tuple( 108 | map(lambda x: x - TO_REMOVE, list(map(int, box))) 109 | ) 110 | 111 | boxes.append(bndbox) 112 | gt_classes.append(self.class_to_ind[name]) 113 | difficult_boxes.append(difficult) 114 | 115 | size = target.find("size") 116 | im_info = tuple(map(int, (size.find("height").text, size.find("width").text))) 117 | 118 | res = { 119 | "boxes": torch.tensor(boxes, dtype=torch.float32), 120 | "labels": torch.tensor(gt_classes), 121 | "difficult": torch.tensor(difficult_boxes), 122 | "im_info": im_info, 123 | } 124 | return res 125 | 126 | def get_img_info(self, index): 127 | img_id = self.ids[index] 128 | anno = ET.parse(self._annopath % img_id).getroot() 129 | size = anno.find("size") 130 | im_info = tuple(map(int, (size.find("height").text, size.find("width").text))) 131 | return {"height": im_info[0], "width": im_info[1]} 132 | 133 | def map_class_id_to_class_name(self, class_id): 134 | return PascalVOCDataset.CLASSES[class_id] 135 | -------------------------------------------------------------------------------- /maskrcnn_benchmark/data/samplers/__init__.py: -------------------------------------------------------------------------------- 1 | # Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved. 2 | from .distributed import DistributedSampler 3 | from .grouped_batch_sampler import GroupedBatchSampler 4 | from .iteration_based_batch_sampler import IterationBasedBatchSampler 5 | 6 | __all__ = ["DistributedSampler", "GroupedBatchSampler", "IterationBasedBatchSampler"] 7 | -------------------------------------------------------------------------------- /maskrcnn_benchmark/data/samplers/distributed.py: -------------------------------------------------------------------------------- 1 | # Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved. 2 | # Code is copy-pasted exactly as in torch.utils.data.distributed. 3 | # FIXME remove this once c10d fixes the bug it has 4 | import math 5 | import torch 6 | import torch.distributed as dist 7 | from torch.utils.data.sampler import Sampler 8 | 9 | 10 | class DistributedSampler(Sampler): 11 | """Sampler that restricts data loading to a subset of the dataset. 12 | It is especially useful in conjunction with 13 | :class:`torch.nn.parallel.DistributedDataParallel`. In such case, each 14 | process can pass a DistributedSampler instance as a DataLoader sampler, 15 | and load a subset of the original dataset that is exclusive to it. 16 | .. note:: 17 | Dataset is assumed to be of constant size. 18 | Arguments: 19 | dataset: Dataset used for sampling. 20 | num_replicas (optional): Number of processes participating in 21 | distributed training. 22 | rank (optional): Rank of the current process within num_replicas. 23 | """ 24 | 25 | def __init__(self, dataset, num_replicas=None, rank=None, shuffle=True): 26 | if num_replicas is None: 27 | if not dist.is_available(): 28 | raise RuntimeError("Requires distributed package to be available") 29 | num_replicas = dist.get_world_size() 30 | if rank is None: 31 | if not dist.is_available(): 32 | raise RuntimeError("Requires distributed package to be available") 33 | rank = dist.get_rank() 34 | self.dataset = dataset 35 | self.num_replicas = num_replicas 36 | self.rank = rank 37 | self.epoch = 0 38 | self.num_samples = int(math.ceil(len(self.dataset) * 1.0 / self.num_replicas)) 39 | self.total_size = self.num_samples * self.num_replicas 40 | self.shuffle = shuffle 41 | 42 | def __iter__(self): 43 | if self.shuffle: 44 | # deterministically shuffle based on epoch 45 | g = torch.Generator() 46 | g.manual_seed(self.epoch) 47 | indices = torch.randperm(len(self.dataset), generator=g).tolist() 48 | else: 49 | indices = torch.arange(len(self.dataset)).tolist() 50 | 51 | # add extra samples to make it evenly divisible 52 | indices += indices[: (self.total_size - len(indices))] 53 | assert len(indices) == self.total_size 54 | 55 | # subsample 56 | offset = self.num_samples * self.rank 57 | indices = indices[offset : offset + self.num_samples] 58 | assert len(indices) == self.num_samples 59 | 60 | return iter(indices) 61 | 62 | def __len__(self): 63 | return self.num_samples 64 | 65 | def set_epoch(self, epoch): 66 | self.epoch = epoch 67 | -------------------------------------------------------------------------------- /maskrcnn_benchmark/data/samplers/iteration_based_batch_sampler.py: -------------------------------------------------------------------------------- 1 | # Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved. 2 | from torch.utils.data.sampler import BatchSampler 3 | 4 | 5 | class IterationBasedBatchSampler(BatchSampler): 6 | """ 7 | Wraps a BatchSampler, resampling from it until 8 | a specified number of iterations have been sampled 9 | """ 10 | 11 | def __init__(self, batch_sampler, num_iterations, start_iter=0): 12 | self.batch_sampler = batch_sampler 13 | self.num_iterations = num_iterations 14 | self.start_iter = start_iter 15 | 16 | def __iter__(self): 17 | iteration = self.start_iter 18 | while iteration <= self.num_iterations: 19 | # if the underlying sampler has a set_epoch method, like 20 | # DistributedSampler, used for making each process see 21 | # a different split of the dataset, then set it 22 | if hasattr(self.batch_sampler.sampler, "set_epoch"): 23 | self.batch_sampler.sampler.set_epoch(iteration) 24 | for batch in self.batch_sampler: 25 | iteration += 1 26 | if iteration > self.num_iterations: 27 | break 28 | yield batch 29 | 30 | def __len__(self): 31 | return self.num_iterations 32 | -------------------------------------------------------------------------------- /maskrcnn_benchmark/data/transforms/__init__.py: -------------------------------------------------------------------------------- 1 | # Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved. 2 | from .transforms import Compose 3 | from .transforms import Resize 4 | from .transforms import RandomHorizontalFlip 5 | from .transforms import ToTensor 6 | from .transforms import Normalize 7 | 8 | from .build import build_transforms 9 | -------------------------------------------------------------------------------- /maskrcnn_benchmark/data/transforms/build.py: -------------------------------------------------------------------------------- 1 | # Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved. 2 | from . import transforms as T 3 | 4 | 5 | def build_transforms(cfg, is_train=True): 6 | if is_train: 7 | min_size = cfg.INPUT.MIN_SIZE_TRAIN 8 | max_size = cfg.INPUT.MAX_SIZE_TRAIN 9 | flip_prob = 0.5 # cfg.INPUT.FLIP_PROB_TRAIN 10 | else: 11 | min_size = cfg.INPUT.MIN_SIZE_TEST 12 | max_size = cfg.INPUT.MAX_SIZE_TEST 13 | flip_prob = 0 14 | 15 | to_bgr255 = cfg.INPUT.TO_BGR255 16 | normalize_transform = T.Normalize( 17 | mean=cfg.INPUT.PIXEL_MEAN, std=cfg.INPUT.PIXEL_STD, to_bgr255=to_bgr255 18 | ) 19 | 20 | transform = T.Compose( 21 | [ 22 | T.Resize(min_size, max_size), 23 | T.RandomHorizontalFlip(flip_prob), 24 | T.ToTensor(), 25 | normalize_transform, 26 | ] 27 | ) 28 | return transform 29 | -------------------------------------------------------------------------------- /maskrcnn_benchmark/data/transforms/transforms.py: -------------------------------------------------------------------------------- 1 | # Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved. 2 | import random 3 | 4 | import torch 5 | import torchvision 6 | from torchvision.transforms import functional as F 7 | 8 | 9 | class Compose(object): 10 | def __init__(self, transforms): 11 | self.transforms = transforms 12 | 13 | def __call__(self, image, target): 14 | for t in self.transforms: 15 | image, target = t(image, target) 16 | return image, target 17 | 18 | def __repr__(self): 19 | format_string = self.__class__.__name__ + "(" 20 | for t in self.transforms: 21 | format_string += "\n" 22 | format_string += " {0}".format(t) 23 | format_string += "\n)" 24 | return format_string 25 | 26 | 27 | class Resize(object): 28 | def __init__(self, min_size, max_size): 29 | if not isinstance(min_size, (list, tuple)): 30 | min_size = (min_size,) 31 | self.min_size = min_size 32 | self.max_size = max_size 33 | 34 | # modified from torchvision to add support for max size 35 | def get_size(self, image_size): 36 | w, h = image_size 37 | size = random.choice(self.min_size) 38 | max_size = self.max_size 39 | if max_size is not None: 40 | min_original_size = float(min((w, h))) 41 | max_original_size = float(max((w, h))) 42 | if max_original_size / min_original_size * size > max_size: 43 | size = int(round(max_size * min_original_size / max_original_size)) 44 | 45 | if (w <= h and w == size) or (h <= w and h == size): 46 | return (h, w) 47 | 48 | if w < h: 49 | ow = size 50 | oh = int(size * h / w) 51 | else: 52 | oh = size 53 | ow = int(size * w / h) 54 | 55 | return (oh, ow) 56 | 57 | def __call__(self, image, target): 58 | size = self.get_size(image.size) 59 | image = F.resize(image, size) 60 | target = target.resize(image.size) 61 | return image, target 62 | 63 | 64 | class RandomHorizontalFlip(object): 65 | def __init__(self, prob=0.5): 66 | self.prob = prob 67 | 68 | def __call__(self, image, target): 69 | if random.random() < self.prob: 70 | image = F.hflip(image) 71 | target = target.transpose(0) 72 | return image, target 73 | 74 | 75 | class ToTensor(object): 76 | def __call__(self, image, target): 77 | return F.to_tensor(image), target 78 | 79 | 80 | class Normalize(object): 81 | def __init__(self, mean, std, to_bgr255=True): 82 | self.mean = mean 83 | self.std = std 84 | self.to_bgr255 = to_bgr255 85 | 86 | def __call__(self, image, target): 87 | if self.to_bgr255: 88 | image = image[[2, 1, 0]] * 255 89 | image = F.normalize(image, mean=self.mean, std=self.std) 90 | return image, target 91 | -------------------------------------------------------------------------------- /maskrcnn_benchmark/engine/__init__.py: -------------------------------------------------------------------------------- 1 | # Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved. 2 | -------------------------------------------------------------------------------- /maskrcnn_benchmark/engine/inference.py: -------------------------------------------------------------------------------- 1 | # Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved. 2 | import datetime 3 | import logging 4 | import time 5 | import os 6 | 7 | import torch 8 | from tqdm import tqdm 9 | 10 | from maskrcnn_benchmark.data.datasets.evaluation import evaluate 11 | from ..utils.comm import is_main_process 12 | from ..utils.comm import all_gather 13 | from ..utils.comm import synchronize 14 | 15 | 16 | def compute_on_dataset(model, data_loader, device): 17 | model.eval() 18 | results_dict = {} 19 | cpu_device = torch.device("cpu") 20 | for i, batch in enumerate(tqdm(data_loader)): 21 | images, targets, image_ids = batch 22 | images = images.to(device) 23 | with torch.no_grad(): 24 | output = model(images) 25 | output = [o.to(cpu_device) for o in output] 26 | results_dict.update( 27 | {img_id: result for img_id, result in zip(image_ids, output)} 28 | ) 29 | return results_dict 30 | 31 | 32 | def _accumulate_predictions_from_multiple_gpus(predictions_per_gpu): 33 | all_predictions = all_gather(predictions_per_gpu) 34 | if not is_main_process(): 35 | return 36 | # merge the list of dicts 37 | predictions = {} 38 | for p in all_predictions: 39 | predictions.update(p) 40 | # convert a dict where the key is the index in a list 41 | image_ids = list(sorted(predictions.keys())) 42 | if len(image_ids) != image_ids[-1] + 1: 43 | logger = logging.getLogger("maskrcnn_benchmark.inference") 44 | logger.warning( 45 | "Number of images that were gathered from multiple processes is not " 46 | "a contiguous set. Some images might be missing from the evaluation" 47 | ) 48 | 49 | # convert to a list 50 | predictions = [predictions[i] for i in image_ids] 51 | return predictions 52 | 53 | 54 | def inference( 55 | model, 56 | data_loader, 57 | dataset_name, 58 | iou_types=("bbox",), 59 | box_only=False, 60 | device="cuda", 61 | expected_results=(), 62 | expected_results_sigma_tol=4, 63 | output_folder=None, 64 | ): 65 | # convert to a torch.device for efficiency 66 | device = torch.device(device) 67 | num_devices = ( 68 | torch.distributed.get_world_size() 69 | if torch.distributed.is_initialized() 70 | else 1 71 | ) 72 | logger = logging.getLogger("maskrcnn_benchmark.inference") 73 | dataset = data_loader.dataset 74 | logger.info("Start evaluation on {} dataset({} images).".format(dataset_name, len(dataset))) 75 | start_time = time.time() 76 | predictions = compute_on_dataset(model, data_loader, device) 77 | # wait for all processes to complete before measuring the time 78 | synchronize() 79 | total_time = time.time() - start_time 80 | total_time_str = str(datetime.timedelta(seconds=total_time)) 81 | logger.info( 82 | "Total inference time: {} ({} s / img per device, on {} devices)".format( 83 | total_time_str, total_time * num_devices / len(dataset), num_devices 84 | ) 85 | ) 86 | 87 | predictions = _accumulate_predictions_from_multiple_gpus(predictions) 88 | if not is_main_process(): 89 | return 90 | 91 | if output_folder: 92 | torch.save(predictions, os.path.join(output_folder, "predictions.pth")) 93 | 94 | extra_args = dict( 95 | box_only=box_only, 96 | iou_types=iou_types, 97 | expected_results=expected_results, 98 | expected_results_sigma_tol=expected_results_sigma_tol, 99 | ) 100 | 101 | return evaluate(dataset=dataset, 102 | predictions=predictions, 103 | output_folder=output_folder, 104 | **extra_args) 105 | -------------------------------------------------------------------------------- /maskrcnn_benchmark/engine/trainer.py: -------------------------------------------------------------------------------- 1 | # Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved. 2 | import datetime 3 | import logging 4 | import time 5 | 6 | import torch 7 | import torch.distributed as dist 8 | 9 | from maskrcnn_benchmark.utils.comm import get_world_size 10 | from maskrcnn_benchmark.utils.metric_logger import MetricLogger 11 | 12 | 13 | def reduce_loss_dict(loss_dict): 14 | """ 15 | Reduce the loss dictionary from all processes so that process with rank 16 | 0 has the averaged results. Returns a dict with the same fields as 17 | loss_dict, after reduction. 18 | """ 19 | world_size = get_world_size() 20 | if world_size < 2: 21 | return loss_dict 22 | with torch.no_grad(): 23 | loss_names = [] 24 | all_losses = [] 25 | for k in sorted(loss_dict.keys()): 26 | loss_names.append(k) 27 | all_losses.append(loss_dict[k]) 28 | all_losses = torch.stack(all_losses, dim=0) 29 | dist.reduce(all_losses, dst=0) 30 | if dist.get_rank() == 0: 31 | # only main process gets accumulated, so only divide by 32 | # world_size in this case 33 | all_losses /= world_size 34 | reduced_losses = {k: v for k, v in zip(loss_names, all_losses)} 35 | return reduced_losses 36 | 37 | 38 | def do_train( 39 | model, 40 | data_loader, 41 | optimizer, 42 | scheduler, 43 | checkpointer, 44 | device, 45 | checkpoint_period, 46 | arguments, 47 | ): 48 | logger = logging.getLogger("maskrcnn_benchmark.trainer") 49 | logger.info("Start training") 50 | meters = MetricLogger(delimiter=" ") 51 | max_iter = len(data_loader) 52 | start_iter = arguments["iteration"] 53 | model.train() 54 | start_training_time = time.time() 55 | end = time.time() 56 | for iteration, (images, targets, _) in enumerate(data_loader, start_iter): 57 | data_time = time.time() - end 58 | iteration = iteration + 1 59 | arguments["iteration"] = iteration 60 | 61 | scheduler.step() 62 | 63 | images = images.to(device) 64 | targets = [target.to(device) for target in targets] 65 | 66 | loss_dict = model(images, targets) 67 | 68 | losses = sum(loss for loss in loss_dict.values()) 69 | 70 | # reduce losses over all GPUs for logging purposes 71 | loss_dict_reduced = reduce_loss_dict(loss_dict) 72 | losses_reduced = sum(loss for loss in loss_dict_reduced.values()) 73 | meters.update(loss=losses_reduced, **loss_dict_reduced) 74 | 75 | optimizer.zero_grad() 76 | losses.backward() 77 | optimizer.step() 78 | 79 | batch_time = time.time() - end 80 | end = time.time() 81 | meters.update(time=batch_time, data=data_time) 82 | 83 | eta_seconds = meters.time.global_avg * (max_iter - iteration) 84 | eta_string = str(datetime.timedelta(seconds=int(eta_seconds))) 85 | 86 | if iteration % 20 == 0 or iteration == max_iter: 87 | logger.info( 88 | meters.delimiter.join( 89 | [ 90 | "eta: {eta}", 91 | "iter: {iter}", 92 | "{meters}", 93 | "lr: {lr:.6f}", 94 | "max mem: {memory:.0f}", 95 | ] 96 | ).format( 97 | eta=eta_string, 98 | iter=iteration, 99 | meters=str(meters), 100 | lr=optimizer.param_groups[0]["lr"], 101 | memory=torch.cuda.max_memory_allocated() / 1024.0 / 1024.0, 102 | ) 103 | ) 104 | if iteration % checkpoint_period == 0: 105 | checkpointer.save("model_{:07d}".format(iteration), **arguments) 106 | if iteration == max_iter: 107 | checkpointer.save("model_final", **arguments) 108 | 109 | total_training_time = time.time() - start_training_time 110 | total_time_str = str(datetime.timedelta(seconds=total_training_time)) 111 | logger.info( 112 | "Total training time: {} ({:.4f} s / it)".format( 113 | total_time_str, total_training_time / (max_iter) 114 | ) 115 | ) 116 | -------------------------------------------------------------------------------- /maskrcnn_benchmark/layers/__init__.py: -------------------------------------------------------------------------------- 1 | # Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved. 2 | import torch 3 | 4 | from .batch_norm import FrozenBatchNorm2d 5 | from .misc import Conv2d 6 | from .misc import ConvTranspose2d 7 | from .misc import BatchNorm2d 8 | from .misc import interpolate 9 | from .nms import nms 10 | from .roi_align import ROIAlign 11 | from .roi_align import roi_align 12 | from .roi_pool import ROIPool 13 | from .roi_pool import roi_pool 14 | from .smooth_l1_loss import smooth_l1_loss 15 | from .sigmoid_focal_loss import SigmoidFocalLoss 16 | 17 | __all__ = ["nms", "roi_align", "ROIAlign", "roi_pool", "ROIPool", 18 | "smooth_l1_loss", "Conv2d", "ConvTranspose2d", "interpolate", 19 | "BatchNorm2d", "FrozenBatchNorm2d", "SigmoidFocalLoss" 20 | ] 21 | 22 | -------------------------------------------------------------------------------- /maskrcnn_benchmark/layers/_utils.py: -------------------------------------------------------------------------------- 1 | # Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved. 2 | import glob 3 | import os.path 4 | 5 | import torch 6 | 7 | try: 8 | from torch.utils.cpp_extension import load as load_ext 9 | from torch.utils.cpp_extension import CUDA_HOME 10 | except ImportError: 11 | raise ImportError("The cpp layer extensions requires PyTorch 0.4 or higher") 12 | 13 | 14 | def _load_C_extensions(): 15 | this_dir = os.path.dirname(os.path.abspath(__file__)) 16 | this_dir = os.path.dirname(this_dir) 17 | this_dir = os.path.join(this_dir, "csrc") 18 | 19 | main_file = glob.glob(os.path.join(this_dir, "*.cpp")) 20 | source_cpu = glob.glob(os.path.join(this_dir, "cpu", "*.cpp")) 21 | source_cuda = glob.glob(os.path.join(this_dir, "cuda", "*.cu")) 22 | 23 | source = main_file + source_cpu 24 | 25 | extra_cflags = [] 26 | if torch.cuda.is_available() and CUDA_HOME is not None: 27 | source.extend(source_cuda) 28 | extra_cflags = ["-DWITH_CUDA"] 29 | source = [os.path.join(this_dir, s) for s in source] 30 | extra_include_paths = [this_dir] 31 | return load_ext( 32 | "torchvision", 33 | source, 34 | extra_cflags=extra_cflags, 35 | extra_include_paths=extra_include_paths, 36 | ) 37 | 38 | 39 | _C = _load_C_extensions() 40 | -------------------------------------------------------------------------------- /maskrcnn_benchmark/layers/batch_norm.py: -------------------------------------------------------------------------------- 1 | # Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved. 2 | import torch 3 | from torch import nn 4 | 5 | 6 | class FrozenBatchNorm2d(nn.Module): 7 | """ 8 | BatchNorm2d where the batch statistics and the affine parameters 9 | are fixed 10 | """ 11 | 12 | def __init__(self, n): 13 | super(FrozenBatchNorm2d, self).__init__() 14 | self.register_buffer("weight", torch.ones(n)) 15 | self.register_buffer("bias", torch.zeros(n)) 16 | self.register_buffer("running_mean", torch.zeros(n)) 17 | self.register_buffer("running_var", torch.ones(n)) 18 | 19 | def forward(self, x): 20 | scale = self.weight * self.running_var.rsqrt() 21 | bias = self.bias - self.running_mean * scale 22 | scale = scale.reshape(1, -1, 1, 1) 23 | bias = bias.reshape(1, -1, 1, 1) 24 | return x * scale + bias 25 | -------------------------------------------------------------------------------- /maskrcnn_benchmark/layers/misc.py: -------------------------------------------------------------------------------- 1 | # Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved. 2 | """ 3 | helper class that supports empty tensors on some nn functions. 4 | 5 | Ideally, add support directly in PyTorch to empty tensors in 6 | those functions. 7 | 8 | This can be removed once https://github.com/pytorch/pytorch/issues/12013 9 | is implemented 10 | """ 11 | 12 | import math 13 | import torch 14 | from torch.nn.modules.utils import _ntuple 15 | 16 | 17 | class _NewEmptyTensorOp(torch.autograd.Function): 18 | @staticmethod 19 | def forward(ctx, x, new_shape): 20 | ctx.shape = x.shape 21 | return x.new_empty(new_shape) 22 | 23 | @staticmethod 24 | def backward(ctx, grad): 25 | shape = ctx.shape 26 | return _NewEmptyTensorOp.apply(grad, shape), None 27 | 28 | 29 | class Conv2d(torch.nn.Conv2d): 30 | def forward(self, x): 31 | if x.numel() > 0: 32 | return super(Conv2d, self).forward(x) 33 | # get output shape 34 | 35 | output_shape = [ 36 | (i + 2 * p - (di * (k - 1) + 1)) // d + 1 37 | for i, p, di, k, d in zip( 38 | x.shape[-2:], self.padding, self.dilation, self.kernel_size, self.stride 39 | ) 40 | ] 41 | output_shape = [x.shape[0], self.weight.shape[0]] + output_shape 42 | return _NewEmptyTensorOp.apply(x, output_shape) 43 | 44 | 45 | class ConvTranspose2d(torch.nn.ConvTranspose2d): 46 | def forward(self, x): 47 | if x.numel() > 0: 48 | return super(ConvTranspose2d, self).forward(x) 49 | # get output shape 50 | 51 | output_shape = [ 52 | (i - 1) * d - 2 * p + (di * (k - 1) + 1) + op 53 | for i, p, di, k, d, op in zip( 54 | x.shape[-2:], 55 | self.padding, 56 | self.dilation, 57 | self.kernel_size, 58 | self.stride, 59 | self.output_padding, 60 | ) 61 | ] 62 | output_shape = [x.shape[0], self.bias.shape[0]] + output_shape 63 | return _NewEmptyTensorOp.apply(x, output_shape) 64 | 65 | 66 | class BatchNorm2d(torch.nn.BatchNorm2d): 67 | def forward(self, x): 68 | if x.numel() > 0: 69 | return super(BatchNorm2d, self).forward(x) 70 | # get output shape 71 | output_shape = x.shape 72 | return _NewEmptyTensorOp.apply(x, output_shape) 73 | 74 | 75 | def interpolate( 76 | input, size=None, scale_factor=None, mode="nearest", align_corners=None 77 | ): 78 | if input.numel() > 0: 79 | return torch.nn.functional.interpolate( 80 | input, size, scale_factor, mode, align_corners 81 | ) 82 | 83 | def _check_size_scale_factor(dim): 84 | if size is None and scale_factor is None: 85 | raise ValueError("either size or scale_factor should be defined") 86 | if size is not None and scale_factor is not None: 87 | raise ValueError("only one of size or scale_factor should be defined") 88 | if ( 89 | scale_factor is not None 90 | and isinstance(scale_factor, tuple) 91 | and len(scale_factor) != dim 92 | ): 93 | raise ValueError( 94 | "scale_factor shape must match input shape. " 95 | "Input is {}D, scale_factor size is {}".format(dim, len(scale_factor)) 96 | ) 97 | 98 | def _output_size(dim): 99 | _check_size_scale_factor(dim) 100 | if size is not None: 101 | return size 102 | scale_factors = _ntuple(dim)(scale_factor) 103 | # math.floor might return float in py2.7 104 | return [ 105 | int(math.floor(input.size(i + 2) * scale_factors[i])) for i in range(dim) 106 | ] 107 | 108 | output_shape = tuple(_output_size(2)) 109 | output_shape = input.shape[:-2] + output_shape 110 | return _NewEmptyTensorOp.apply(input, output_shape) 111 | -------------------------------------------------------------------------------- /maskrcnn_benchmark/layers/nms.py: -------------------------------------------------------------------------------- 1 | # Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved. 2 | # from ._utils import _C 3 | from maskrcnn_benchmark import _C 4 | 5 | nms = _C.nms 6 | # nms.__doc__ = """ 7 | # This function performs Non-maximum suppresion""" 8 | -------------------------------------------------------------------------------- /maskrcnn_benchmark/layers/roi_align.py: -------------------------------------------------------------------------------- 1 | # Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved. 2 | import torch 3 | from torch import nn 4 | from torch.autograd import Function 5 | from torch.autograd.function import once_differentiable 6 | from torch.nn.modules.utils import _pair 7 | 8 | from maskrcnn_benchmark import _C 9 | 10 | 11 | class _ROIAlign(Function): 12 | @staticmethod 13 | def forward(ctx, input, roi, output_size, spatial_scale, sampling_ratio): 14 | ctx.save_for_backward(roi) 15 | ctx.output_size = _pair(output_size) 16 | ctx.spatial_scale = spatial_scale 17 | ctx.sampling_ratio = sampling_ratio 18 | ctx.input_shape = input.size() 19 | output = _C.roi_align_forward( 20 | input, roi, spatial_scale, output_size[0], output_size[1], sampling_ratio 21 | ) 22 | return output 23 | 24 | @staticmethod 25 | @once_differentiable 26 | def backward(ctx, grad_output): 27 | rois, = ctx.saved_tensors 28 | output_size = ctx.output_size 29 | spatial_scale = ctx.spatial_scale 30 | sampling_ratio = ctx.sampling_ratio 31 | bs, ch, h, w = ctx.input_shape 32 | grad_input = _C.roi_align_backward( 33 | grad_output, 34 | rois, 35 | spatial_scale, 36 | output_size[0], 37 | output_size[1], 38 | bs, 39 | ch, 40 | h, 41 | w, 42 | sampling_ratio, 43 | ) 44 | return grad_input, None, None, None, None 45 | 46 | 47 | roi_align = _ROIAlign.apply 48 | 49 | 50 | class ROIAlign(nn.Module): 51 | def __init__(self, output_size, spatial_scale, sampling_ratio): 52 | super(ROIAlign, self).__init__() 53 | self.output_size = output_size 54 | self.spatial_scale = spatial_scale 55 | self.sampling_ratio = sampling_ratio 56 | 57 | def forward(self, input, rois): 58 | return roi_align( 59 | input, rois, self.output_size, self.spatial_scale, self.sampling_ratio 60 | ) 61 | 62 | def __repr__(self): 63 | tmpstr = self.__class__.__name__ + "(" 64 | tmpstr += "output_size=" + str(self.output_size) 65 | tmpstr += ", spatial_scale=" + str(self.spatial_scale) 66 | tmpstr += ", sampling_ratio=" + str(self.sampling_ratio) 67 | tmpstr += ")" 68 | return tmpstr 69 | -------------------------------------------------------------------------------- /maskrcnn_benchmark/layers/roi_pool.py: -------------------------------------------------------------------------------- 1 | # Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved. 2 | import torch 3 | from torch import nn 4 | from torch.autograd import Function 5 | from torch.autograd.function import once_differentiable 6 | from torch.nn.modules.utils import _pair 7 | 8 | from maskrcnn_benchmark import _C 9 | 10 | 11 | class _ROIPool(Function): 12 | @staticmethod 13 | def forward(ctx, input, roi, output_size, spatial_scale): 14 | ctx.output_size = _pair(output_size) 15 | ctx.spatial_scale = spatial_scale 16 | ctx.input_shape = input.size() 17 | output, argmax = _C.roi_pool_forward( 18 | input, roi, spatial_scale, output_size[0], output_size[1] 19 | ) 20 | ctx.save_for_backward(input, roi, argmax) 21 | return output 22 | 23 | @staticmethod 24 | @once_differentiable 25 | def backward(ctx, grad_output): 26 | input, rois, argmax = ctx.saved_tensors 27 | output_size = ctx.output_size 28 | spatial_scale = ctx.spatial_scale 29 | bs, ch, h, w = ctx.input_shape 30 | grad_input = _C.roi_pool_backward( 31 | grad_output, 32 | input, 33 | rois, 34 | argmax, 35 | spatial_scale, 36 | output_size[0], 37 | output_size[1], 38 | bs, 39 | ch, 40 | h, 41 | w, 42 | ) 43 | return grad_input, None, None, None 44 | 45 | 46 | roi_pool = _ROIPool.apply 47 | 48 | 49 | class ROIPool(nn.Module): 50 | def __init__(self, output_size, spatial_scale): 51 | super(ROIPool, self).__init__() 52 | self.output_size = output_size 53 | self.spatial_scale = spatial_scale 54 | 55 | def forward(self, input, rois): 56 | return roi_pool(input, rois, self.output_size, self.spatial_scale) 57 | 58 | def __repr__(self): 59 | tmpstr = self.__class__.__name__ + "(" 60 | tmpstr += "output_size=" + str(self.output_size) 61 | tmpstr += ", spatial_scale=" + str(self.spatial_scale) 62 | tmpstr += ")" 63 | return tmpstr 64 | -------------------------------------------------------------------------------- /maskrcnn_benchmark/layers/sigmoid_focal_loss.py: -------------------------------------------------------------------------------- 1 | import torch 2 | from torch import nn 3 | from torch.autograd import Function 4 | from torch.autograd.function import once_differentiable 5 | 6 | from maskrcnn_benchmark import _C 7 | 8 | # TODO: Use JIT to replace CUDA implementation in the future. 9 | class _SigmoidFocalLoss(Function): 10 | @staticmethod 11 | def forward(ctx, logits, targets, gamma, alpha): 12 | ctx.save_for_backward(logits, targets) 13 | num_classes = logits.shape[1] 14 | ctx.num_classes = num_classes 15 | ctx.gamma = gamma 16 | ctx.alpha = alpha 17 | 18 | losses = _C.sigmoid_focalloss_forward( 19 | logits, targets, num_classes, gamma, alpha 20 | ) 21 | return losses 22 | 23 | @staticmethod 24 | @once_differentiable 25 | def backward(ctx, d_loss): 26 | logits, targets = ctx.saved_tensors 27 | num_classes = ctx.num_classes 28 | gamma = ctx.gamma 29 | alpha = ctx.alpha 30 | d_loss = d_loss.contiguous() 31 | d_logits = _C.sigmoid_focalloss_backward( 32 | logits, targets, d_loss, num_classes, gamma, alpha 33 | ) 34 | return d_logits, None, None, None, None 35 | 36 | 37 | sigmoid_focal_loss_cuda = _SigmoidFocalLoss.apply 38 | 39 | 40 | def sigmoid_focal_loss_cpu(logits, targets, gamma, alpha): 41 | num_classes = logits.shape[1] 42 | gamma = gamma[0] 43 | alpha = alpha[0] 44 | dtype = targets.dtype 45 | device = targets.device 46 | class_range = torch.arange(1, num_classes+1, dtype=dtype, device=device).unsqueeze(0) 47 | 48 | t = targets.unsqueeze(1) 49 | p = torch.sigmoid(logits) 50 | term1 = (1 - p) ** gamma * torch.log(p) 51 | term2 = p ** gamma * torch.log(1 - p) 52 | return -(t == class_range).float() * term1 * alpha - ((t != class_range) * (t >= 0)).float() * term2 * (1 - alpha) 53 | 54 | 55 | class SigmoidFocalLoss(nn.Module): 56 | def __init__(self, gamma, alpha): 57 | super(SigmoidFocalLoss, self).__init__() 58 | self.gamma = gamma 59 | self.alpha = alpha 60 | 61 | def forward(self, logits, targets): 62 | device = logits.device 63 | if logits.is_cuda: 64 | loss_func = sigmoid_focal_loss_cuda 65 | else: 66 | loss_func = sigmoid_focal_loss_cpu 67 | 68 | loss = loss_func(logits, targets, self.gamma, self.alpha) 69 | return loss.sum() 70 | 71 | def __repr__(self): 72 | tmpstr = self.__class__.__name__ + "(" 73 | tmpstr += "gamma=" + str(self.gamma) 74 | tmpstr += ", alpha=" + str(self.alpha) 75 | tmpstr += ")" 76 | return tmpstr 77 | -------------------------------------------------------------------------------- /maskrcnn_benchmark/layers/smooth_l1_loss.py: -------------------------------------------------------------------------------- 1 | # Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved. 2 | import torch 3 | 4 | 5 | # TODO maybe push this to nn? 6 | def smooth_l1_loss(input, target, beta=1. / 9, size_average=True): 7 | """ 8 | very similar to the smooth_l1_loss from pytorch, but with 9 | the extra beta parameter 10 | """ 11 | n = torch.abs(input - target) 12 | cond = n < beta 13 | loss = torch.where(cond, 0.5 * n ** 2 / beta, n - 0.5 * beta) 14 | if size_average: 15 | return loss.mean() 16 | return loss.sum() 17 | -------------------------------------------------------------------------------- /maskrcnn_benchmark/modeling/__init__.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Res2Net/Res2Net-maskrcnn/dd50e0f7b767333335e18fb23054fe27e6b40519/maskrcnn_benchmark/modeling/__init__.py -------------------------------------------------------------------------------- /maskrcnn_benchmark/modeling/backbone/__init__.py: -------------------------------------------------------------------------------- 1 | # Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved. 2 | from .backbone import build_backbone 3 | from . import fbnet 4 | -------------------------------------------------------------------------------- /maskrcnn_benchmark/modeling/backbone/backbone.py: -------------------------------------------------------------------------------- 1 | # Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved. 2 | from collections import OrderedDict 3 | 4 | from torch import nn 5 | 6 | from maskrcnn_benchmark.modeling import registry 7 | from maskrcnn_benchmark.modeling.make_layers import conv_with_kaiming_uniform 8 | from . import fpn as fpn_module 9 | from . import resnet 10 | from . import res2net_builder 11 | 12 | @registry.BACKBONES.register("R-50-C4") 13 | @registry.BACKBONES.register("R-50-C5") 14 | @registry.BACKBONES.register("R-101-C4") 15 | @registry.BACKBONES.register("R-101-C5") 16 | def build_resnet_backbone(cfg): 17 | body = resnet.ResNet(cfg) 18 | model = nn.Sequential(OrderedDict([("body", body)])) 19 | model.out_channels = cfg.MODEL.RESNETS.BACKBONE_OUT_CHANNELS 20 | return model 21 | 22 | 23 | @registry.BACKBONES.register("R-50-FPN") 24 | @registry.BACKBONES.register("R-101-FPN") 25 | @registry.BACKBONES.register("R-152-FPN") 26 | def build_resnet_fpn_backbone(cfg): 27 | body = resnet.ResNet(cfg) 28 | in_channels_stage2 = cfg.MODEL.RESNETS.RES2_OUT_CHANNELS 29 | out_channels = cfg.MODEL.RESNETS.BACKBONE_OUT_CHANNELS 30 | fpn = fpn_module.FPN( 31 | in_channels_list=[ 32 | in_channels_stage2, 33 | in_channels_stage2 * 2, 34 | in_channels_stage2 * 4, 35 | in_channels_stage2 * 8, 36 | ], 37 | out_channels=out_channels, 38 | conv_block=conv_with_kaiming_uniform( 39 | cfg.MODEL.FPN.USE_GN, cfg.MODEL.FPN.USE_RELU 40 | ), 41 | top_blocks=fpn_module.LastLevelMaxPool(), 42 | ) 43 | model = nn.Sequential(OrderedDict([("body", body), ("fpn", fpn)])) 44 | model.out_channels = out_channels 45 | return model 46 | 47 | 48 | @registry.BACKBONES.register("R-50-FPN-RETINANET") 49 | @registry.BACKBONES.register("R-101-FPN-RETINANET") 50 | def build_resnet_fpn_p3p7_backbone(cfg): 51 | body = resnet.ResNet(cfg) 52 | in_channels_stage2 = cfg.MODEL.RESNETS.RES2_OUT_CHANNELS 53 | out_channels = cfg.MODEL.RESNETS.BACKBONE_OUT_CHANNELS 54 | in_channels_p6p7 = in_channels_stage2 * 8 if cfg.MODEL.RETINANET.USE_C5 \ 55 | else out_channels 56 | fpn = fpn_module.FPN( 57 | in_channels_list=[ 58 | 0, 59 | in_channels_stage2 * 2, 60 | in_channels_stage2 * 4, 61 | in_channels_stage2 * 8, 62 | ], 63 | out_channels=out_channels, 64 | conv_block=conv_with_kaiming_uniform( 65 | cfg.MODEL.FPN.USE_GN, cfg.MODEL.FPN.USE_RELU 66 | ), 67 | top_blocks=fpn_module.LastLevelP6P7(in_channels_p6p7, out_channels), 68 | ) 69 | model = nn.Sequential(OrderedDict([("body", body), ("fpn", fpn)])) 70 | model.out_channels = out_channels 71 | return model 72 | 73 | 74 | def build_backbone(cfg): 75 | assert cfg.MODEL.BACKBONE.CONV_BODY in registry.BACKBONES, \ 76 | "cfg.MODEL.BACKBONE.CONV_BODY: {} are not registered in registry".format( 77 | cfg.MODEL.BACKBONE.CONV_BODY 78 | ) 79 | return registry.BACKBONES[cfg.MODEL.BACKBONE.CONV_BODY](cfg) 80 | -------------------------------------------------------------------------------- /maskrcnn_benchmark/modeling/backbone/fbnet_modeldef.py: -------------------------------------------------------------------------------- 1 | from __future__ import absolute_import, division, print_function, unicode_literals 2 | 3 | 4 | def add_archs(archs): 5 | global MODEL_ARCH 6 | for x in archs: 7 | assert x not in MODEL_ARCH, "Duplicated model name {} existed".format(x) 8 | MODEL_ARCH[x] = archs[x] 9 | 10 | 11 | MODEL_ARCH = { 12 | "default": { 13 | "block_op_type": [ 14 | # stage 0 15 | ["ir_k3"], 16 | # stage 1 17 | ["ir_k3"] * 2, 18 | # stage 2 19 | ["ir_k3"] * 3, 20 | # stage 3 21 | ["ir_k3"] * 7, 22 | # stage 4, bbox head 23 | ["ir_k3"] * 4, 24 | # stage 5, rpn 25 | ["ir_k3"] * 3, 26 | # stage 5, mask head 27 | ["ir_k3"] * 5, 28 | ], 29 | "block_cfg": { 30 | "first": [32, 2], 31 | "stages": [ 32 | # [t, c, n, s] 33 | # stage 0 34 | [[1, 16, 1, 1]], 35 | # stage 1 36 | [[6, 24, 2, 2]], 37 | # stage 2 38 | [[6, 32, 3, 2]], 39 | # stage 3 40 | [[6, 64, 4, 2], [6, 96, 3, 1]], 41 | # stage 4, bbox head 42 | [[4, 160, 1, 2], [6, 160, 2, 1], [6, 240, 1, 1]], 43 | # [[6, 160, 3, 2], [6, 320, 1, 1]], 44 | # stage 5, rpn head 45 | [[6, 96, 3, 1]], 46 | # stage 6, mask head 47 | [[4, 160, 1, 1], [6, 160, 3, 1], [3, 80, 1, -2]], 48 | ], 49 | # [c, channel_scale] 50 | "last": [1280, 0.0], 51 | "backbone": [0, 1, 2, 3], 52 | "rpn": [5], 53 | "bbox": [4], 54 | "mask": [6], 55 | }, 56 | }, 57 | "xirb16d_dsmask": { 58 | "block_op_type": [ 59 | # stage 0 60 | ["ir_k3"], 61 | # stage 1 62 | ["ir_k3"] * 2, 63 | # stage 2 64 | ["ir_k3"] * 3, 65 | # stage 3 66 | ["ir_k3"] * 7, 67 | # stage 4, bbox head 68 | ["ir_k3"] * 4, 69 | # stage 5, mask head 70 | ["ir_k3"] * 5, 71 | # stage 6, rpn 72 | ["ir_k3"] * 3, 73 | ], 74 | "block_cfg": { 75 | "first": [16, 2], 76 | "stages": [ 77 | # [t, c, n, s] 78 | # stage 0 79 | [[1, 16, 1, 1]], 80 | # stage 1 81 | [[6, 32, 2, 2]], 82 | # stage 2 83 | [[6, 48, 3, 2]], 84 | # stage 3 85 | [[6, 96, 4, 2], [6, 128, 3, 1]], 86 | # stage 4, bbox head 87 | [[4, 128, 1, 2], [6, 128, 2, 1], [6, 160, 1, 1]], 88 | # stage 5, mask head 89 | [[4, 128, 1, 2], [6, 128, 2, 1], [6, 128, 1, -2], [3, 64, 1, -2]], 90 | # stage 6, rpn head 91 | [[6, 128, 3, 1]], 92 | ], 93 | # [c, channel_scale] 94 | "last": [1280, 0.0], 95 | "backbone": [0, 1, 2, 3], 96 | "rpn": [6], 97 | "bbox": [4], 98 | "mask": [5], 99 | }, 100 | }, 101 | "mobilenet_v2": { 102 | "block_op_type": [ 103 | # stage 0 104 | ["ir_k3"], 105 | # stage 1 106 | ["ir_k3"] * 2, 107 | # stage 2 108 | ["ir_k3"] * 3, 109 | # stage 3 110 | ["ir_k3"] * 7, 111 | # stage 4 112 | ["ir_k3"] * 4, 113 | ], 114 | "block_cfg": { 115 | "first": [32, 2], 116 | "stages": [ 117 | # [t, c, n, s] 118 | # stage 0 119 | [[1, 16, 1, 1]], 120 | # stage 1 121 | [[6, 24, 2, 2]], 122 | # stage 2 123 | [[6, 32, 3, 2]], 124 | # stage 3 125 | [[6, 64, 4, 2], [6, 96, 3, 1]], 126 | # stage 4 127 | [[6, 160, 3, 1], [6, 320, 1, 1]], 128 | ], 129 | # [c, channel_scale] 130 | "last": [1280, 0.0], 131 | "backbone": [0, 1, 2, 3], 132 | "bbox": [4], 133 | }, 134 | }, 135 | } 136 | -------------------------------------------------------------------------------- /maskrcnn_benchmark/modeling/backbone/fpn.py: -------------------------------------------------------------------------------- 1 | # Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved. 2 | import torch 3 | import torch.nn.functional as F 4 | from torch import nn 5 | 6 | 7 | class FPN(nn.Module): 8 | """ 9 | Module that adds FPN on top of a list of feature maps. 10 | The feature maps are currently supposed to be in increasing depth 11 | order, and must be consecutive 12 | """ 13 | 14 | def __init__( 15 | self, in_channels_list, out_channels, conv_block, top_blocks=None 16 | ): 17 | """ 18 | Arguments: 19 | in_channels_list (list[int]): number of channels for each feature map that 20 | will be fed 21 | out_channels (int): number of channels of the FPN representation 22 | top_blocks (nn.Module or None): if provided, an extra operation will 23 | be performed on the output of the last (smallest resolution) 24 | FPN output, and the result will extend the result list 25 | """ 26 | super(FPN, self).__init__() 27 | self.inner_blocks = [] 28 | self.layer_blocks = [] 29 | for idx, in_channels in enumerate(in_channels_list, 1): 30 | inner_block = "fpn_inner{}".format(idx) 31 | layer_block = "fpn_layer{}".format(idx) 32 | 33 | if in_channels == 0: 34 | continue 35 | inner_block_module = conv_block(in_channels, out_channels, 1) 36 | layer_block_module = conv_block(out_channels, out_channels, 3, 1) 37 | self.add_module(inner_block, inner_block_module) 38 | self.add_module(layer_block, layer_block_module) 39 | self.inner_blocks.append(inner_block) 40 | self.layer_blocks.append(layer_block) 41 | self.top_blocks = top_blocks 42 | 43 | def forward(self, x): 44 | """ 45 | Arguments: 46 | x (list[Tensor]): feature maps for each feature level. 47 | Returns: 48 | results (tuple[Tensor]): feature maps after FPN layers. 49 | They are ordered from highest resolution first. 50 | """ 51 | last_inner = getattr(self, self.inner_blocks[-1])(x[-1]) 52 | results = [] 53 | results.append(getattr(self, self.layer_blocks[-1])(last_inner)) 54 | for feature, inner_block, layer_block in zip( 55 | x[:-1][::-1], self.inner_blocks[:-1][::-1], self.layer_blocks[:-1][::-1] 56 | ): 57 | if not inner_block: 58 | continue 59 | inner_top_down = F.interpolate(last_inner, scale_factor=2, mode="nearest") 60 | inner_lateral = getattr(self, inner_block)(feature) 61 | # TODO use size instead of scale to make it robust to different sizes 62 | # inner_top_down = F.upsample(last_inner, size=inner_lateral.shape[-2:], 63 | # mode='bilinear', align_corners=False) 64 | last_inner = inner_lateral + inner_top_down 65 | results.insert(0, getattr(self, layer_block)(last_inner)) 66 | 67 | if isinstance(self.top_blocks, LastLevelP6P7): 68 | last_results = self.top_blocks(x[-1], results[-1]) 69 | results.extend(last_results) 70 | elif isinstance(self.top_blocks, LastLevelMaxPool): 71 | last_results = self.top_blocks(results[-1]) 72 | results.extend(last_results) 73 | 74 | return tuple(results) 75 | 76 | 77 | class LastLevelMaxPool(nn.Module): 78 | def forward(self, x): 79 | return [F.max_pool2d(x, 1, 2, 0)] 80 | 81 | 82 | class LastLevelP6P7(nn.Module): 83 | """ 84 | This module is used in RetinaNet to generate extra layers, P6 and P7. 85 | """ 86 | def __init__(self, in_channels, out_channels): 87 | super(LastLevelP6P7, self).__init__() 88 | self.p6 = nn.Conv2d(in_channels, out_channels, 3, 2, 1) 89 | self.p7 = nn.Conv2d(out_channels, out_channels, 3, 2, 1) 90 | for module in [self.p6, self.p7]: 91 | nn.init.kaiming_uniform_(module.weight, a=1) 92 | nn.init.constant_(module.bias, 0) 93 | self.use_P5 = in_channels == out_channels 94 | 95 | def forward(self, c5, p5): 96 | x = p5 if self.use_P5 else c5 97 | p6 = self.p6(x) 98 | p7 = self.p7(F.relu(p6)) 99 | return [p6, p7] 100 | -------------------------------------------------------------------------------- /maskrcnn_benchmark/modeling/backbone/res2net_builder.py: -------------------------------------------------------------------------------- 1 | # Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved. 2 | from collections import OrderedDict 3 | 4 | from torch import nn 5 | 6 | from maskrcnn_benchmark.modeling import registry 7 | from maskrcnn_benchmark.modeling.make_layers import conv_with_kaiming_uniform 8 | from . import fpn as fpn_module 9 | from . import res2net 10 | 11 | 12 | @registry.BACKBONES.register("R2-50-C4") 13 | @registry.BACKBONES.register("R2-50-C5") 14 | @registry.BACKBONES.register("R2-101-C4") 15 | @registry.BACKBONES.register("R2-101-C5") 16 | def build_res2net_backbone(cfg): 17 | body = res2net.Res2Net(cfg) 18 | model = nn.Sequential(OrderedDict([("body", body)])) 19 | model.out_channels = cfg.MODEL.RESNETS.BACKBONE_OUT_CHANNELS 20 | return model 21 | 22 | 23 | @registry.BACKBONES.register("R2-50-FPN") 24 | @registry.BACKBONES.register("R2-101-FPN") 25 | @registry.BACKBONES.register("R2-152-FPN") 26 | def build_res2net_fpn_backbone(cfg): 27 | body = res2net.Res2Net(cfg) 28 | in_channels_stage2 = cfg.MODEL.RESNETS.RES2_OUT_CHANNELS 29 | out_channels = cfg.MODEL.RESNETS.BACKBONE_OUT_CHANNELS 30 | fpn = fpn_module.FPN( 31 | in_channels_list=[ 32 | in_channels_stage2, 33 | in_channels_stage2 * 2, 34 | in_channels_stage2 * 4, 35 | in_channels_stage2 * 8, 36 | ], 37 | out_channels=out_channels, 38 | conv_block=conv_with_kaiming_uniform( 39 | cfg.MODEL.FPN.USE_GN, cfg.MODEL.FPN.USE_RELU 40 | ), 41 | top_blocks=fpn_module.LastLevelMaxPool(), 42 | ) 43 | model = nn.Sequential(OrderedDict([("body", body), ("fpn", fpn)])) 44 | model.out_channels = out_channels 45 | return model 46 | 47 | 48 | @registry.BACKBONES.register("R2-50-FPN-RETINANET") 49 | @registry.BACKBONES.register("R2-101-FPN-RETINANET") 50 | def build_res2net_fpn_p3p7_backbone(cfg): 51 | body = res2net.Res2Net(cfg) 52 | in_channels_stage2 = cfg.MODEL.RESNETS.RES2_OUT_CHANNELS 53 | out_channels = cfg.MODEL.RESNETS.BACKBONE_OUT_CHANNELS 54 | in_channels_p6p7 = in_channels_stage2 * 8 if cfg.MODEL.RETINANET.USE_C5 \ 55 | else out_channels 56 | fpn = fpn_module.FPN( 57 | in_channels_list=[ 58 | 0, 59 | in_channels_stage2 * 2, 60 | in_channels_stage2 * 4, 61 | in_channels_stage2 * 8, 62 | ], 63 | out_channels=out_channels, 64 | conv_block=conv_with_kaiming_uniform( 65 | cfg.MODEL.FPN.USE_GN, cfg.MODEL.FPN.USE_RELU 66 | ), 67 | top_blocks=fpn_module.LastLevelP6P7(in_channels_p6p7, out_channels), 68 | ) 69 | model = nn.Sequential(OrderedDict([("body", body), ("fpn", fpn)])) 70 | model.out_channels = out_channels 71 | return model 72 | 73 | 74 | # def build_backbone(cfg): 75 | # assert cfg.MODEL.BACKBONE.CONV_BODY in registry.BACKBONES, \ 76 | # "cfg.MODEL.BACKBONE.CONV_BODY: {} are not registered in registry".format( 77 | # cfg.MODEL.BACKBONE.CONV_BODY 78 | # ) 79 | # return registry.BACKBONES[cfg.MODEL.BACKBONE.CONV_BODY](cfg) 80 | -------------------------------------------------------------------------------- /maskrcnn_benchmark/modeling/balanced_positive_negative_sampler.py: -------------------------------------------------------------------------------- 1 | # Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved. 2 | import torch 3 | 4 | 5 | class BalancedPositiveNegativeSampler(object): 6 | """ 7 | This class samples batches, ensuring that they contain a fixed proportion of positives 8 | """ 9 | 10 | def __init__(self, batch_size_per_image, positive_fraction): 11 | """ 12 | Arguments: 13 | batch_size_per_image (int): number of elements to be selected per image 14 | positive_fraction (float): percentace of positive elements per batch 15 | """ 16 | self.batch_size_per_image = batch_size_per_image 17 | self.positive_fraction = positive_fraction 18 | 19 | def __call__(self, matched_idxs): 20 | """ 21 | Arguments: 22 | matched idxs: list of tensors containing -1, 0 or positive values. 23 | Each tensor corresponds to a specific image. 24 | -1 values are ignored, 0 are considered as negatives and > 0 as 25 | positives. 26 | 27 | Returns: 28 | pos_idx (list[tensor]) 29 | neg_idx (list[tensor]) 30 | 31 | Returns two lists of binary masks for each image. 32 | The first list contains the positive elements that were selected, 33 | and the second list the negative example. 34 | """ 35 | pos_idx = [] 36 | neg_idx = [] 37 | for matched_idxs_per_image in matched_idxs: 38 | positive = torch.nonzero(matched_idxs_per_image >= 1).squeeze(1) 39 | negative = torch.nonzero(matched_idxs_per_image == 0).squeeze(1) 40 | 41 | num_pos = int(self.batch_size_per_image * self.positive_fraction) 42 | # protect against not enough positive examples 43 | num_pos = min(positive.numel(), num_pos) 44 | num_neg = self.batch_size_per_image - num_pos 45 | # protect against not enough negative examples 46 | num_neg = min(negative.numel(), num_neg) 47 | 48 | # randomly select positive and negative examples 49 | perm1 = torch.randperm(positive.numel(), device=positive.device)[:num_pos] 50 | perm2 = torch.randperm(negative.numel(), device=negative.device)[:num_neg] 51 | 52 | pos_idx_per_image = positive[perm1] 53 | neg_idx_per_image = negative[perm2] 54 | 55 | # create binary mask from indices 56 | pos_idx_per_image_mask = torch.zeros_like( 57 | matched_idxs_per_image, dtype=torch.uint8 58 | ) 59 | neg_idx_per_image_mask = torch.zeros_like( 60 | matched_idxs_per_image, dtype=torch.uint8 61 | ) 62 | pos_idx_per_image_mask[pos_idx_per_image] = 1 63 | neg_idx_per_image_mask[neg_idx_per_image] = 1 64 | 65 | pos_idx.append(pos_idx_per_image_mask) 66 | neg_idx.append(neg_idx_per_image_mask) 67 | 68 | return pos_idx, neg_idx 69 | -------------------------------------------------------------------------------- /maskrcnn_benchmark/modeling/box_coder.py: -------------------------------------------------------------------------------- 1 | # Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved. 2 | import math 3 | 4 | import torch 5 | 6 | 7 | class BoxCoder(object): 8 | """ 9 | This class encodes and decodes a set of bounding boxes into 10 | the representation used for training the regressors. 11 | """ 12 | 13 | def __init__(self, weights, bbox_xform_clip=math.log(1000. / 16)): 14 | """ 15 | Arguments: 16 | weights (4-element tuple) 17 | bbox_xform_clip (float) 18 | """ 19 | self.weights = weights 20 | self.bbox_xform_clip = bbox_xform_clip 21 | 22 | def encode(self, reference_boxes, proposals): 23 | """ 24 | Encode a set of proposals with respect to some 25 | reference boxes 26 | 27 | Arguments: 28 | reference_boxes (Tensor): reference boxes 29 | proposals (Tensor): boxes to be encoded 30 | """ 31 | 32 | TO_REMOVE = 1 # TODO remove 33 | ex_widths = proposals[:, 2] - proposals[:, 0] + TO_REMOVE 34 | ex_heights = proposals[:, 3] - proposals[:, 1] + TO_REMOVE 35 | ex_ctr_x = proposals[:, 0] + 0.5 * ex_widths 36 | ex_ctr_y = proposals[:, 1] + 0.5 * ex_heights 37 | 38 | gt_widths = reference_boxes[:, 2] - reference_boxes[:, 0] + TO_REMOVE 39 | gt_heights = reference_boxes[:, 3] - reference_boxes[:, 1] + TO_REMOVE 40 | gt_ctr_x = reference_boxes[:, 0] + 0.5 * gt_widths 41 | gt_ctr_y = reference_boxes[:, 1] + 0.5 * gt_heights 42 | 43 | wx, wy, ww, wh = self.weights 44 | targets_dx = wx * (gt_ctr_x - ex_ctr_x) / ex_widths 45 | targets_dy = wy * (gt_ctr_y - ex_ctr_y) / ex_heights 46 | targets_dw = ww * torch.log(gt_widths / ex_widths) 47 | targets_dh = wh * torch.log(gt_heights / ex_heights) 48 | 49 | targets = torch.stack((targets_dx, targets_dy, targets_dw, targets_dh), dim=1) 50 | return targets 51 | 52 | def decode(self, rel_codes, boxes): 53 | """ 54 | From a set of original boxes and encoded relative box offsets, 55 | get the decoded boxes. 56 | 57 | Arguments: 58 | rel_codes (Tensor): encoded boxes 59 | boxes (Tensor): reference boxes. 60 | """ 61 | 62 | boxes = boxes.to(rel_codes.dtype) 63 | 64 | TO_REMOVE = 1 # TODO remove 65 | widths = boxes[:, 2] - boxes[:, 0] + TO_REMOVE 66 | heights = boxes[:, 3] - boxes[:, 1] + TO_REMOVE 67 | ctr_x = boxes[:, 0] + 0.5 * widths 68 | ctr_y = boxes[:, 1] + 0.5 * heights 69 | 70 | wx, wy, ww, wh = self.weights 71 | dx = rel_codes[:, 0::4] / wx 72 | dy = rel_codes[:, 1::4] / wy 73 | dw = rel_codes[:, 2::4] / ww 74 | dh = rel_codes[:, 3::4] / wh 75 | 76 | # Prevent sending too large values into torch.exp() 77 | dw = torch.clamp(dw, max=self.bbox_xform_clip) 78 | dh = torch.clamp(dh, max=self.bbox_xform_clip) 79 | 80 | pred_ctr_x = dx * widths[:, None] + ctr_x[:, None] 81 | pred_ctr_y = dy * heights[:, None] + ctr_y[:, None] 82 | pred_w = torch.exp(dw) * widths[:, None] 83 | pred_h = torch.exp(dh) * heights[:, None] 84 | 85 | pred_boxes = torch.zeros_like(rel_codes) 86 | # x1 87 | pred_boxes[:, 0::4] = pred_ctr_x - 0.5 * pred_w 88 | # y1 89 | pred_boxes[:, 1::4] = pred_ctr_y - 0.5 * pred_h 90 | # x2 (note: "- 1" is correct; don't be fooled by the asymmetry) 91 | pred_boxes[:, 2::4] = pred_ctr_x + 0.5 * pred_w - 1 92 | # y2 (note: "- 1" is correct; don't be fooled by the asymmetry) 93 | pred_boxes[:, 3::4] = pred_ctr_y + 0.5 * pred_h - 1 94 | 95 | return pred_boxes 96 | -------------------------------------------------------------------------------- /maskrcnn_benchmark/modeling/detector/__init__.py: -------------------------------------------------------------------------------- 1 | # Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved. 2 | from .detectors import build_detection_model 3 | -------------------------------------------------------------------------------- /maskrcnn_benchmark/modeling/detector/detectors.py: -------------------------------------------------------------------------------- 1 | # Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved. 2 | from .generalized_rcnn import GeneralizedRCNN 3 | 4 | 5 | _DETECTION_META_ARCHITECTURES = {"GeneralizedRCNN": GeneralizedRCNN} 6 | 7 | 8 | def build_detection_model(cfg): 9 | meta_arch = _DETECTION_META_ARCHITECTURES[cfg.MODEL.META_ARCHITECTURE] 10 | return meta_arch(cfg) 11 | -------------------------------------------------------------------------------- /maskrcnn_benchmark/modeling/detector/generalized_rcnn.py: -------------------------------------------------------------------------------- 1 | # Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved. 2 | """ 3 | Implements the Generalized R-CNN framework 4 | """ 5 | 6 | import torch 7 | from torch import nn 8 | 9 | from maskrcnn_benchmark.structures.image_list import to_image_list 10 | 11 | from ..backbone import build_backbone 12 | from ..rpn.rpn import build_rpn 13 | from ..roi_heads.roi_heads import build_roi_heads 14 | 15 | 16 | class GeneralizedRCNN(nn.Module): 17 | """ 18 | Main class for Generalized R-CNN. Currently supports boxes and masks. 19 | It consists of three main parts: 20 | - backbone 21 | - rpn 22 | - heads: takes the features + the proposals from the RPN and computes 23 | detections / masks from it. 24 | """ 25 | 26 | def __init__(self, cfg): 27 | super(GeneralizedRCNN, self).__init__() 28 | 29 | self.backbone = build_backbone(cfg) 30 | self.rpn = build_rpn(cfg, self.backbone.out_channels) 31 | self.roi_heads = build_roi_heads(cfg, self.backbone.out_channels) 32 | 33 | def forward(self, images, targets=None): 34 | """ 35 | Arguments: 36 | images (list[Tensor] or ImageList): images to be processed 37 | targets (list[BoxList]): ground-truth boxes present in the image (optional) 38 | 39 | Returns: 40 | result (list[BoxList] or dict[Tensor]): the output from the model. 41 | During training, it returns a dict[Tensor] which contains the losses. 42 | During testing, it returns list[BoxList] contains additional fields 43 | like `scores`, `labels` and `mask` (for Mask R-CNN models). 44 | 45 | """ 46 | if self.training and targets is None: 47 | raise ValueError("In training mode, targets should be passed") 48 | images = to_image_list(images) 49 | features = self.backbone(images.tensors) 50 | proposals, proposal_losses = self.rpn(images, features, targets) 51 | if self.roi_heads: 52 | x, result, detector_losses = self.roi_heads(features, proposals, targets) 53 | else: 54 | # RPN-only models don't have roi_heads 55 | x = features 56 | result = proposals 57 | detector_losses = {} 58 | 59 | if self.training: 60 | losses = {} 61 | losses.update(detector_losses) 62 | losses.update(proposal_losses) 63 | return losses 64 | 65 | return result 66 | -------------------------------------------------------------------------------- /maskrcnn_benchmark/modeling/make_layers.py: -------------------------------------------------------------------------------- 1 | # Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved. 2 | """ 3 | Miscellaneous utility functions 4 | """ 5 | 6 | import torch 7 | from torch import nn 8 | from torch.nn import functional as F 9 | from maskrcnn_benchmark.config import cfg 10 | from maskrcnn_benchmark.layers import Conv2d 11 | from maskrcnn_benchmark.modeling.poolers import Pooler 12 | 13 | 14 | def get_group_gn(dim, dim_per_gp, num_groups): 15 | """get number of groups used by GroupNorm, based on number of channels.""" 16 | assert dim_per_gp == -1 or num_groups == -1, \ 17 | "GroupNorm: can only specify G or C/G." 18 | 19 | if dim_per_gp > 0: 20 | assert dim % dim_per_gp == 0, \ 21 | "dim: {}, dim_per_gp: {}".format(dim, dim_per_gp) 22 | group_gn = dim // dim_per_gp 23 | else: 24 | assert dim % num_groups == 0, \ 25 | "dim: {}, num_groups: {}".format(dim, num_groups) 26 | group_gn = num_groups 27 | 28 | return group_gn 29 | 30 | 31 | def group_norm(out_channels, affine=True, divisor=1): 32 | out_channels = out_channels // divisor 33 | dim_per_gp = cfg.MODEL.GROUP_NORM.DIM_PER_GP // divisor 34 | num_groups = cfg.MODEL.GROUP_NORM.NUM_GROUPS // divisor 35 | eps = cfg.MODEL.GROUP_NORM.EPSILON # default: 1e-5 36 | return torch.nn.GroupNorm( 37 | get_group_gn(out_channels, dim_per_gp, num_groups), 38 | out_channels, 39 | eps, 40 | affine 41 | ) 42 | 43 | 44 | def make_conv3x3( 45 | in_channels, 46 | out_channels, 47 | dilation=1, 48 | stride=1, 49 | use_gn=False, 50 | use_relu=False, 51 | kaiming_init=True 52 | ): 53 | conv = Conv2d( 54 | in_channels, 55 | out_channels, 56 | kernel_size=3, 57 | stride=stride, 58 | padding=dilation, 59 | dilation=dilation, 60 | bias=False if use_gn else True 61 | ) 62 | if kaiming_init: 63 | nn.init.kaiming_normal_( 64 | conv.weight, mode="fan_out", nonlinearity="relu" 65 | ) 66 | else: 67 | torch.nn.init.normal_(conv.weight, std=0.01) 68 | if not use_gn: 69 | nn.init.constant_(conv.bias, 0) 70 | module = [conv,] 71 | if use_gn: 72 | module.append(group_norm(out_channels)) 73 | if use_relu: 74 | module.append(nn.ReLU(inplace=True)) 75 | if len(module) > 1: 76 | return nn.Sequential(*module) 77 | return conv 78 | 79 | 80 | def make_fc(dim_in, hidden_dim, use_gn=False): 81 | ''' 82 | Caffe2 implementation uses XavierFill, which in fact 83 | corresponds to kaiming_uniform_ in PyTorch 84 | ''' 85 | if use_gn: 86 | fc = nn.Linear(dim_in, hidden_dim, bias=False) 87 | nn.init.kaiming_uniform_(fc.weight, a=1) 88 | return nn.Sequential(fc, group_norm(hidden_dim)) 89 | fc = nn.Linear(dim_in, hidden_dim) 90 | nn.init.kaiming_uniform_(fc.weight, a=1) 91 | nn.init.constant_(fc.bias, 0) 92 | return fc 93 | 94 | 95 | def conv_with_kaiming_uniform(use_gn=False, use_relu=False): 96 | def make_conv( 97 | in_channels, out_channels, kernel_size, stride=1, dilation=1 98 | ): 99 | conv = Conv2d( 100 | in_channels, 101 | out_channels, 102 | kernel_size=kernel_size, 103 | stride=stride, 104 | padding=dilation * (kernel_size - 1) // 2, 105 | dilation=dilation, 106 | bias=False if use_gn else True 107 | ) 108 | # Caffe2 implementation uses XavierFill, which in fact 109 | # corresponds to kaiming_uniform_ in PyTorch 110 | nn.init.kaiming_uniform_(conv.weight, a=1) 111 | if not use_gn: 112 | nn.init.constant_(conv.bias, 0) 113 | module = [conv,] 114 | if use_gn: 115 | module.append(group_norm(out_channels)) 116 | if use_relu: 117 | module.append(nn.ReLU(inplace=True)) 118 | if len(module) > 1: 119 | return nn.Sequential(*module) 120 | return conv 121 | 122 | return make_conv 123 | -------------------------------------------------------------------------------- /maskrcnn_benchmark/modeling/registry.py: -------------------------------------------------------------------------------- 1 | # Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved. 2 | 3 | from maskrcnn_benchmark.utils.registry import Registry 4 | 5 | BACKBONES = Registry() 6 | RPN_HEADS = Registry() 7 | ROI_BOX_FEATURE_EXTRACTORS = Registry() 8 | ROI_BOX_PREDICTOR = Registry() 9 | ROI_KEYPOINT_FEATURE_EXTRACTORS = Registry() 10 | ROI_KEYPOINT_PREDICTOR = Registry() 11 | ROI_MASK_FEATURE_EXTRACTORS = Registry() 12 | ROI_MASK_PREDICTOR = Registry() 13 | -------------------------------------------------------------------------------- /maskrcnn_benchmark/modeling/roi_heads/__init__.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Res2Net/Res2Net-maskrcnn/dd50e0f7b767333335e18fb23054fe27e6b40519/maskrcnn_benchmark/modeling/roi_heads/__init__.py -------------------------------------------------------------------------------- /maskrcnn_benchmark/modeling/roi_heads/box_head/__init__.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Res2Net/Res2Net-maskrcnn/dd50e0f7b767333335e18fb23054fe27e6b40519/maskrcnn_benchmark/modeling/roi_heads/box_head/__init__.py -------------------------------------------------------------------------------- /maskrcnn_benchmark/modeling/roi_heads/box_head/box_head.py: -------------------------------------------------------------------------------- 1 | # Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved. 2 | import torch 3 | from torch import nn 4 | 5 | from .roi_box_feature_extractors import make_roi_box_feature_extractor 6 | from .roi_box_predictors import make_roi_box_predictor 7 | from .inference import make_roi_box_post_processor 8 | from .loss import make_roi_box_loss_evaluator 9 | 10 | 11 | class ROIBoxHead(torch.nn.Module): 12 | """ 13 | Generic Box Head class. 14 | """ 15 | 16 | def __init__(self, cfg, in_channels): 17 | super(ROIBoxHead, self).__init__() 18 | self.feature_extractor = make_roi_box_feature_extractor(cfg, in_channels) 19 | self.predictor = make_roi_box_predictor( 20 | cfg, self.feature_extractor.out_channels) 21 | self.post_processor = make_roi_box_post_processor(cfg) 22 | self.loss_evaluator = make_roi_box_loss_evaluator(cfg) 23 | 24 | def forward(self, features, proposals, targets=None): 25 | """ 26 | Arguments: 27 | features (list[Tensor]): feature-maps from possibly several levels 28 | proposals (list[BoxList]): proposal boxes 29 | targets (list[BoxList], optional): the ground-truth targets. 30 | 31 | Returns: 32 | x (Tensor): the result of the feature extractor 33 | proposals (list[BoxList]): during training, the subsampled proposals 34 | are returned. During testing, the predicted boxlists are returned 35 | losses (dict[Tensor]): During training, returns the losses for the 36 | head. During testing, returns an empty dict. 37 | """ 38 | 39 | if self.training: 40 | # Faster R-CNN subsamples during training the proposals with a fixed 41 | # positive / negative ratio 42 | with torch.no_grad(): 43 | proposals = self.loss_evaluator.subsample(proposals, targets) 44 | 45 | # extract features that will be fed to the final classifier. The 46 | # feature_extractor generally corresponds to the pooler + heads 47 | x = self.feature_extractor(features, proposals) 48 | # final classifier that converts the features into predictions 49 | class_logits, box_regression = self.predictor(x) 50 | 51 | if not self.training: 52 | result = self.post_processor((class_logits, box_regression), proposals) 53 | return x, result, {} 54 | 55 | loss_classifier, loss_box_reg = self.loss_evaluator( 56 | [class_logits], [box_regression] 57 | ) 58 | return ( 59 | x, 60 | proposals, 61 | dict(loss_classifier=loss_classifier, loss_box_reg=loss_box_reg), 62 | ) 63 | 64 | 65 | def build_roi_box_head(cfg, in_channels): 66 | """ 67 | Constructs a new box head. 68 | By default, uses ROIBoxHead, but if it turns out not to be enough, just register a new class 69 | and make it a parameter in the config 70 | """ 71 | return ROIBoxHead(cfg, in_channels) 72 | -------------------------------------------------------------------------------- /maskrcnn_benchmark/modeling/roi_heads/box_head/roi_box_predictors.py: -------------------------------------------------------------------------------- 1 | # Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved. 2 | from maskrcnn_benchmark.modeling import registry 3 | from torch import nn 4 | 5 | 6 | @registry.ROI_BOX_PREDICTOR.register("FastRCNNPredictor") 7 | class FastRCNNPredictor(nn.Module): 8 | def __init__(self, config, in_channels): 9 | super(FastRCNNPredictor, self).__init__() 10 | assert in_channels is not None 11 | 12 | num_inputs = in_channels 13 | 14 | num_classes = config.MODEL.ROI_BOX_HEAD.NUM_CLASSES 15 | self.avgpool = nn.AdaptiveAvgPool2d(1) 16 | self.cls_score = nn.Linear(num_inputs, num_classes) 17 | num_bbox_reg_classes = 2 if config.MODEL.CLS_AGNOSTIC_BBOX_REG else num_classes 18 | self.bbox_pred = nn.Linear(num_inputs, num_bbox_reg_classes * 4) 19 | 20 | nn.init.normal_(self.cls_score.weight, mean=0, std=0.01) 21 | nn.init.constant_(self.cls_score.bias, 0) 22 | 23 | nn.init.normal_(self.bbox_pred.weight, mean=0, std=0.001) 24 | nn.init.constant_(self.bbox_pred.bias, 0) 25 | 26 | def forward(self, x): 27 | x = self.avgpool(x) 28 | x = x.view(x.size(0), -1) 29 | cls_logit = self.cls_score(x) 30 | bbox_pred = self.bbox_pred(x) 31 | return cls_logit, bbox_pred 32 | 33 | 34 | @registry.ROI_BOX_PREDICTOR.register("FPNPredictor") 35 | class FPNPredictor(nn.Module): 36 | def __init__(self, cfg, in_channels): 37 | super(FPNPredictor, self).__init__() 38 | num_classes = cfg.MODEL.ROI_BOX_HEAD.NUM_CLASSES 39 | representation_size = in_channels 40 | 41 | self.cls_score = nn.Linear(representation_size, num_classes) 42 | num_bbox_reg_classes = 2 if cfg.MODEL.CLS_AGNOSTIC_BBOX_REG else num_classes 43 | self.bbox_pred = nn.Linear(representation_size, num_bbox_reg_classes * 4) 44 | 45 | nn.init.normal_(self.cls_score.weight, std=0.01) 46 | nn.init.normal_(self.bbox_pred.weight, std=0.001) 47 | for l in [self.cls_score, self.bbox_pred]: 48 | nn.init.constant_(l.bias, 0) 49 | 50 | def forward(self, x): 51 | if x.ndimension() == 4: 52 | assert list(x.shape[2:]) == [1, 1] 53 | x = x.view(x.size(0), -1) 54 | scores = self.cls_score(x) 55 | bbox_deltas = self.bbox_pred(x) 56 | 57 | return scores, bbox_deltas 58 | 59 | 60 | def make_roi_box_predictor(cfg, in_channels): 61 | func = registry.ROI_BOX_PREDICTOR[cfg.MODEL.ROI_BOX_HEAD.PREDICTOR] 62 | return func(cfg, in_channels) 63 | -------------------------------------------------------------------------------- /maskrcnn_benchmark/modeling/roi_heads/keypoint_head/__init__.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Res2Net/Res2Net-maskrcnn/dd50e0f7b767333335e18fb23054fe27e6b40519/maskrcnn_benchmark/modeling/roi_heads/keypoint_head/__init__.py -------------------------------------------------------------------------------- /maskrcnn_benchmark/modeling/roi_heads/keypoint_head/keypoint_head.py: -------------------------------------------------------------------------------- 1 | import torch 2 | 3 | from .roi_keypoint_feature_extractors import make_roi_keypoint_feature_extractor 4 | from .roi_keypoint_predictors import make_roi_keypoint_predictor 5 | from .inference import make_roi_keypoint_post_processor 6 | from .loss import make_roi_keypoint_loss_evaluator 7 | 8 | 9 | class ROIKeypointHead(torch.nn.Module): 10 | def __init__(self, cfg, in_channels): 11 | super(ROIKeypointHead, self).__init__() 12 | self.cfg = cfg.clone() 13 | self.feature_extractor = make_roi_keypoint_feature_extractor(cfg, in_channels) 14 | self.predictor = make_roi_keypoint_predictor( 15 | cfg, self.feature_extractor.out_channels) 16 | self.post_processor = make_roi_keypoint_post_processor(cfg) 17 | self.loss_evaluator = make_roi_keypoint_loss_evaluator(cfg) 18 | 19 | def forward(self, features, proposals, targets=None): 20 | """ 21 | Arguments: 22 | features (list[Tensor]): feature-maps from possibly several levels 23 | proposals (list[BoxList]): proposal boxes 24 | targets (list[BoxList], optional): the ground-truth targets. 25 | 26 | Returns: 27 | x (Tensor): the result of the feature extractor 28 | proposals (list[BoxList]): during training, the original proposals 29 | are returned. During testing, the predicted boxlists are returned 30 | with the `mask` field set 31 | losses (dict[Tensor]): During training, returns the losses for the 32 | head. During testing, returns an empty dict. 33 | """ 34 | if self.training: 35 | with torch.no_grad(): 36 | proposals = self.loss_evaluator.subsample(proposals, targets) 37 | 38 | x = self.feature_extractor(features, proposals) 39 | kp_logits = self.predictor(x) 40 | 41 | if not self.training: 42 | result = self.post_processor(kp_logits, proposals) 43 | return x, result, {} 44 | 45 | loss_kp = self.loss_evaluator(proposals, kp_logits) 46 | 47 | return x, proposals, dict(loss_kp=loss_kp) 48 | 49 | 50 | def build_roi_keypoint_head(cfg, in_channels): 51 | return ROIKeypointHead(cfg, in_channels) 52 | -------------------------------------------------------------------------------- /maskrcnn_benchmark/modeling/roi_heads/keypoint_head/roi_keypoint_feature_extractors.py: -------------------------------------------------------------------------------- 1 | from torch import nn 2 | from torch.nn import functional as F 3 | 4 | from maskrcnn_benchmark.modeling import registry 5 | from maskrcnn_benchmark.modeling.poolers import Pooler 6 | 7 | from maskrcnn_benchmark.layers import Conv2d 8 | 9 | 10 | @registry.ROI_KEYPOINT_FEATURE_EXTRACTORS.register("KeypointRCNNFeatureExtractor") 11 | class KeypointRCNNFeatureExtractor(nn.Module): 12 | def __init__(self, cfg, in_channels): 13 | super(KeypointRCNNFeatureExtractor, self).__init__() 14 | 15 | resolution = cfg.MODEL.ROI_KEYPOINT_HEAD.POOLER_RESOLUTION 16 | scales = cfg.MODEL.ROI_KEYPOINT_HEAD.POOLER_SCALES 17 | sampling_ratio = cfg.MODEL.ROI_KEYPOINT_HEAD.POOLER_SAMPLING_RATIO 18 | pooler = Pooler( 19 | output_size=(resolution, resolution), 20 | scales=scales, 21 | sampling_ratio=sampling_ratio, 22 | ) 23 | self.pooler = pooler 24 | 25 | input_features = in_channels 26 | layers = cfg.MODEL.ROI_KEYPOINT_HEAD.CONV_LAYERS 27 | next_feature = input_features 28 | self.blocks = [] 29 | for layer_idx, layer_features in enumerate(layers, 1): 30 | layer_name = "conv_fcn{}".format(layer_idx) 31 | module = Conv2d(next_feature, layer_features, 3, stride=1, padding=1) 32 | nn.init.kaiming_normal_(module.weight, mode="fan_out", nonlinearity="relu") 33 | nn.init.constant_(module.bias, 0) 34 | self.add_module(layer_name, module) 35 | next_feature = layer_features 36 | self.blocks.append(layer_name) 37 | self.out_channels = layer_features 38 | 39 | def forward(self, x, proposals): 40 | x = self.pooler(x, proposals) 41 | for layer_name in self.blocks: 42 | x = F.relu(getattr(self, layer_name)(x)) 43 | return x 44 | 45 | 46 | def make_roi_keypoint_feature_extractor(cfg, in_channels): 47 | func = registry.ROI_KEYPOINT_FEATURE_EXTRACTORS[ 48 | cfg.MODEL.ROI_KEYPOINT_HEAD.FEATURE_EXTRACTOR 49 | ] 50 | return func(cfg, in_channels) 51 | -------------------------------------------------------------------------------- /maskrcnn_benchmark/modeling/roi_heads/keypoint_head/roi_keypoint_predictors.py: -------------------------------------------------------------------------------- 1 | from torch import nn 2 | 3 | from maskrcnn_benchmark import layers 4 | from maskrcnn_benchmark.modeling import registry 5 | 6 | 7 | @registry.ROI_KEYPOINT_PREDICTOR.register("KeypointRCNNPredictor") 8 | class KeypointRCNNPredictor(nn.Module): 9 | def __init__(self, cfg, in_channels): 10 | super(KeypointRCNNPredictor, self).__init__() 11 | input_features = in_channels 12 | num_keypoints = cfg.MODEL.ROI_KEYPOINT_HEAD.NUM_CLASSES 13 | deconv_kernel = 4 14 | self.kps_score_lowres = layers.ConvTranspose2d( 15 | input_features, 16 | num_keypoints, 17 | deconv_kernel, 18 | stride=2, 19 | padding=deconv_kernel // 2 - 1, 20 | ) 21 | nn.init.kaiming_normal_( 22 | self.kps_score_lowres.weight, mode="fan_out", nonlinearity="relu" 23 | ) 24 | nn.init.constant_(self.kps_score_lowres.bias, 0) 25 | self.up_scale = 2 26 | self.out_channels = num_keypoints 27 | 28 | def forward(self, x): 29 | x = self.kps_score_lowres(x) 30 | x = layers.interpolate( 31 | x, scale_factor=self.up_scale, mode="bilinear", align_corners=False 32 | ) 33 | return x 34 | 35 | 36 | def make_roi_keypoint_predictor(cfg, in_channels): 37 | func = registry.ROI_KEYPOINT_PREDICTOR[cfg.MODEL.ROI_KEYPOINT_HEAD.PREDICTOR] 38 | return func(cfg, in_channels) 39 | -------------------------------------------------------------------------------- /maskrcnn_benchmark/modeling/roi_heads/mask_head/__init__.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Res2Net/Res2Net-maskrcnn/dd50e0f7b767333335e18fb23054fe27e6b40519/maskrcnn_benchmark/modeling/roi_heads/mask_head/__init__.py -------------------------------------------------------------------------------- /maskrcnn_benchmark/modeling/roi_heads/mask_head/mask_head.py: -------------------------------------------------------------------------------- 1 | # Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved. 2 | import torch 3 | from torch import nn 4 | 5 | from maskrcnn_benchmark.structures.bounding_box import BoxList 6 | 7 | from .roi_mask_feature_extractors import make_roi_mask_feature_extractor 8 | from .roi_mask_predictors import make_roi_mask_predictor 9 | from .inference import make_roi_mask_post_processor 10 | from .loss import make_roi_mask_loss_evaluator 11 | 12 | 13 | def keep_only_positive_boxes(boxes): 14 | """ 15 | Given a set of BoxList containing the `labels` field, 16 | return a set of BoxList for which `labels > 0`. 17 | 18 | Arguments: 19 | boxes (list of BoxList) 20 | """ 21 | assert isinstance(boxes, (list, tuple)) 22 | assert isinstance(boxes[0], BoxList) 23 | assert boxes[0].has_field("labels") 24 | positive_boxes = [] 25 | positive_inds = [] 26 | num_boxes = 0 27 | for boxes_per_image in boxes: 28 | labels = boxes_per_image.get_field("labels") 29 | inds_mask = labels > 0 30 | inds = inds_mask.nonzero().squeeze(1) 31 | positive_boxes.append(boxes_per_image[inds]) 32 | positive_inds.append(inds_mask) 33 | return positive_boxes, positive_inds 34 | 35 | 36 | class ROIMaskHead(torch.nn.Module): 37 | def __init__(self, cfg, in_channels): 38 | super(ROIMaskHead, self).__init__() 39 | self.cfg = cfg.clone() 40 | self.feature_extractor = make_roi_mask_feature_extractor(cfg, in_channels) 41 | self.predictor = make_roi_mask_predictor( 42 | cfg, self.feature_extractor.out_channels) 43 | self.post_processor = make_roi_mask_post_processor(cfg) 44 | self.loss_evaluator = make_roi_mask_loss_evaluator(cfg) 45 | 46 | def forward(self, features, proposals, targets=None): 47 | """ 48 | Arguments: 49 | features (list[Tensor]): feature-maps from possibly several levels 50 | proposals (list[BoxList]): proposal boxes 51 | targets (list[BoxList], optional): the ground-truth targets. 52 | 53 | Returns: 54 | x (Tensor): the result of the feature extractor 55 | proposals (list[BoxList]): during training, the original proposals 56 | are returned. During testing, the predicted boxlists are returned 57 | with the `mask` field set 58 | losses (dict[Tensor]): During training, returns the losses for the 59 | head. During testing, returns an empty dict. 60 | """ 61 | 62 | if self.training: 63 | # during training, only focus on positive boxes 64 | all_proposals = proposals 65 | proposals, positive_inds = keep_only_positive_boxes(proposals) 66 | if self.training and self.cfg.MODEL.ROI_MASK_HEAD.SHARE_BOX_FEATURE_EXTRACTOR: 67 | x = features 68 | x = x[torch.cat(positive_inds, dim=0)] 69 | else: 70 | x = self.feature_extractor(features, proposals) 71 | mask_logits = self.predictor(x) 72 | 73 | if not self.training: 74 | result = self.post_processor(mask_logits, proposals) 75 | return x, result, {} 76 | 77 | loss_mask = self.loss_evaluator(proposals, mask_logits, targets) 78 | 79 | return x, all_proposals, dict(loss_mask=loss_mask) 80 | 81 | 82 | def build_roi_mask_head(cfg, in_channels): 83 | return ROIMaskHead(cfg, in_channels) 84 | -------------------------------------------------------------------------------- /maskrcnn_benchmark/modeling/roi_heads/mask_head/roi_mask_feature_extractors.py: -------------------------------------------------------------------------------- 1 | # Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved. 2 | from torch import nn 3 | from torch.nn import functional as F 4 | 5 | from ..box_head.roi_box_feature_extractors import ResNet50Conv5ROIFeatureExtractor 6 | from maskrcnn_benchmark.modeling import registry 7 | from maskrcnn_benchmark.modeling.poolers import Pooler 8 | from maskrcnn_benchmark.modeling.make_layers import make_conv3x3 9 | 10 | 11 | registry.ROI_MASK_FEATURE_EXTRACTORS.register( 12 | "ResNet50Conv5ROIFeatureExtractor", ResNet50Conv5ROIFeatureExtractor 13 | ) 14 | 15 | 16 | @registry.ROI_MASK_FEATURE_EXTRACTORS.register("MaskRCNNFPNFeatureExtractor") 17 | class MaskRCNNFPNFeatureExtractor(nn.Module): 18 | """ 19 | Heads for FPN for classification 20 | """ 21 | 22 | def __init__(self, cfg, in_channels): 23 | """ 24 | Arguments: 25 | num_classes (int): number of output classes 26 | input_size (int): number of channels of the input once it's flattened 27 | representation_size (int): size of the intermediate representation 28 | """ 29 | super(MaskRCNNFPNFeatureExtractor, self).__init__() 30 | 31 | resolution = cfg.MODEL.ROI_MASK_HEAD.POOLER_RESOLUTION 32 | scales = cfg.MODEL.ROI_MASK_HEAD.POOLER_SCALES 33 | sampling_ratio = cfg.MODEL.ROI_MASK_HEAD.POOLER_SAMPLING_RATIO 34 | pooler = Pooler( 35 | output_size=(resolution, resolution), 36 | scales=scales, 37 | sampling_ratio=sampling_ratio, 38 | ) 39 | input_size = in_channels 40 | self.pooler = pooler 41 | 42 | use_gn = cfg.MODEL.ROI_MASK_HEAD.USE_GN 43 | layers = cfg.MODEL.ROI_MASK_HEAD.CONV_LAYERS 44 | dilation = cfg.MODEL.ROI_MASK_HEAD.DILATION 45 | 46 | next_feature = input_size 47 | self.blocks = [] 48 | for layer_idx, layer_features in enumerate(layers, 1): 49 | layer_name = "mask_fcn{}".format(layer_idx) 50 | module = make_conv3x3( 51 | next_feature, layer_features, 52 | dilation=dilation, stride=1, use_gn=use_gn 53 | ) 54 | self.add_module(layer_name, module) 55 | next_feature = layer_features 56 | self.blocks.append(layer_name) 57 | self.out_channels = layer_features 58 | 59 | def forward(self, x, proposals): 60 | x = self.pooler(x, proposals) 61 | 62 | for layer_name in self.blocks: 63 | x = F.relu(getattr(self, layer_name)(x)) 64 | 65 | return x 66 | 67 | 68 | def make_roi_mask_feature_extractor(cfg, in_channels): 69 | func = registry.ROI_MASK_FEATURE_EXTRACTORS[ 70 | cfg.MODEL.ROI_MASK_HEAD.FEATURE_EXTRACTOR 71 | ] 72 | return func(cfg, in_channels) 73 | -------------------------------------------------------------------------------- /maskrcnn_benchmark/modeling/roi_heads/mask_head/roi_mask_predictors.py: -------------------------------------------------------------------------------- 1 | # Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved. 2 | from torch import nn 3 | from torch.nn import functional as F 4 | 5 | from maskrcnn_benchmark.layers import Conv2d 6 | from maskrcnn_benchmark.layers import ConvTranspose2d 7 | from maskrcnn_benchmark.modeling import registry 8 | 9 | 10 | @registry.ROI_MASK_PREDICTOR.register("MaskRCNNC4Predictor") 11 | class MaskRCNNC4Predictor(nn.Module): 12 | def __init__(self, cfg, in_channels): 13 | super(MaskRCNNC4Predictor, self).__init__() 14 | num_classes = cfg.MODEL.ROI_BOX_HEAD.NUM_CLASSES 15 | dim_reduced = cfg.MODEL.ROI_MASK_HEAD.CONV_LAYERS[-1] 16 | num_inputs = in_channels 17 | 18 | self.conv5_mask = ConvTranspose2d(num_inputs, dim_reduced, 2, 2, 0) 19 | self.mask_fcn_logits = Conv2d(dim_reduced, num_classes, 1, 1, 0) 20 | 21 | for name, param in self.named_parameters(): 22 | if "bias" in name: 23 | nn.init.constant_(param, 0) 24 | elif "weight" in name: 25 | # Caffe2 implementation uses MSRAFill, which in fact 26 | # corresponds to kaiming_normal_ in PyTorch 27 | nn.init.kaiming_normal_(param, mode="fan_out", nonlinearity="relu") 28 | 29 | def forward(self, x): 30 | x = F.relu(self.conv5_mask(x)) 31 | return self.mask_fcn_logits(x) 32 | 33 | 34 | @registry.ROI_MASK_PREDICTOR.register("MaskRCNNConv1x1Predictor") 35 | class MaskRCNNConv1x1Predictor(nn.Module): 36 | def __init__(self, cfg, in_channels): 37 | super(MaskRCNNConv1x1Predictor, self).__init__() 38 | num_classes = cfg.MODEL.ROI_BOX_HEAD.NUM_CLASSES 39 | num_inputs = in_channels 40 | 41 | self.mask_fcn_logits = Conv2d(num_inputs, num_classes, 1, 1, 0) 42 | 43 | for name, param in self.named_parameters(): 44 | if "bias" in name: 45 | nn.init.constant_(param, 0) 46 | elif "weight" in name: 47 | # Caffe2 implementation uses MSRAFill, which in fact 48 | # corresponds to kaiming_normal_ in PyTorch 49 | nn.init.kaiming_normal_(param, mode="fan_out", nonlinearity="relu") 50 | 51 | def forward(self, x): 52 | return self.mask_fcn_logits(x) 53 | 54 | 55 | def make_roi_mask_predictor(cfg, in_channels): 56 | func = registry.ROI_MASK_PREDICTOR[cfg.MODEL.ROI_MASK_HEAD.PREDICTOR] 57 | return func(cfg, in_channels) 58 | -------------------------------------------------------------------------------- /maskrcnn_benchmark/modeling/roi_heads/roi_heads.py: -------------------------------------------------------------------------------- 1 | # Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved. 2 | import torch 3 | 4 | from .box_head.box_head import build_roi_box_head 5 | from .mask_head.mask_head import build_roi_mask_head 6 | from .keypoint_head.keypoint_head import build_roi_keypoint_head 7 | 8 | 9 | class CombinedROIHeads(torch.nn.ModuleDict): 10 | """ 11 | Combines a set of individual heads (for box prediction or masks) into a single 12 | head. 13 | """ 14 | 15 | def __init__(self, cfg, heads): 16 | super(CombinedROIHeads, self).__init__(heads) 17 | self.cfg = cfg.clone() 18 | if cfg.MODEL.MASK_ON and cfg.MODEL.ROI_MASK_HEAD.SHARE_BOX_FEATURE_EXTRACTOR: 19 | self.mask.feature_extractor = self.box.feature_extractor 20 | if cfg.MODEL.KEYPOINT_ON and cfg.MODEL.ROI_KEYPOINT_HEAD.SHARE_BOX_FEATURE_EXTRACTOR: 21 | self.keypoint.feature_extractor = self.box.feature_extractor 22 | 23 | def forward(self, features, proposals, targets=None): 24 | losses = {} 25 | # TODO rename x to roi_box_features, if it doesn't increase memory consumption 26 | x, detections, loss_box = self.box(features, proposals, targets) 27 | losses.update(loss_box) 28 | if self.cfg.MODEL.MASK_ON: 29 | mask_features = features 30 | # optimization: during training, if we share the feature extractor between 31 | # the box and the mask heads, then we can reuse the features already computed 32 | if ( 33 | self.training 34 | and self.cfg.MODEL.ROI_MASK_HEAD.SHARE_BOX_FEATURE_EXTRACTOR 35 | ): 36 | mask_features = x 37 | # During training, self.box() will return the unaltered proposals as "detections" 38 | # this makes the API consistent during training and testing 39 | x, detections, loss_mask = self.mask(mask_features, detections, targets) 40 | losses.update(loss_mask) 41 | 42 | if self.cfg.MODEL.KEYPOINT_ON: 43 | keypoint_features = features 44 | # optimization: during training, if we share the feature extractor between 45 | # the box and the mask heads, then we can reuse the features already computed 46 | if ( 47 | self.training 48 | and self.cfg.MODEL.ROI_KEYPOINT_HEAD.SHARE_BOX_FEATURE_EXTRACTOR 49 | ): 50 | keypoint_features = x 51 | # During training, self.box() will return the unaltered proposals as "detections" 52 | # this makes the API consistent during training and testing 53 | x, detections, loss_keypoint = self.keypoint(keypoint_features, detections, targets) 54 | losses.update(loss_keypoint) 55 | return x, detections, losses 56 | 57 | 58 | def build_roi_heads(cfg, in_channels): 59 | # individually create the heads, that will be combined together 60 | # afterwards 61 | roi_heads = [] 62 | if cfg.MODEL.RETINANET_ON: 63 | return [] 64 | 65 | if not cfg.MODEL.RPN_ONLY: 66 | roi_heads.append(("box", build_roi_box_head(cfg, in_channels))) 67 | if cfg.MODEL.MASK_ON: 68 | roi_heads.append(("mask", build_roi_mask_head(cfg, in_channels))) 69 | if cfg.MODEL.KEYPOINT_ON: 70 | roi_heads.append(("keypoint", build_roi_keypoint_head(cfg, in_channels))) 71 | 72 | # combine individual heads in a single module 73 | if roi_heads: 74 | roi_heads = CombinedROIHeads(cfg, roi_heads) 75 | 76 | return roi_heads 77 | -------------------------------------------------------------------------------- /maskrcnn_benchmark/modeling/rpn/__init__.py: -------------------------------------------------------------------------------- 1 | # Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved. 2 | # from .rpn import build_rpn 3 | -------------------------------------------------------------------------------- /maskrcnn_benchmark/modeling/rpn/retinanet/__init__.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Res2Net/Res2Net-maskrcnn/dd50e0f7b767333335e18fb23054fe27e6b40519/maskrcnn_benchmark/modeling/rpn/retinanet/__init__.py -------------------------------------------------------------------------------- /maskrcnn_benchmark/modeling/rpn/retinanet/loss.py: -------------------------------------------------------------------------------- 1 | """ 2 | This file contains specific functions for computing losses on the RetinaNet 3 | file 4 | """ 5 | 6 | import torch 7 | from torch.nn import functional as F 8 | 9 | from ..utils import concat_box_prediction_layers 10 | 11 | from maskrcnn_benchmark.layers import smooth_l1_loss 12 | from maskrcnn_benchmark.layers import SigmoidFocalLoss 13 | from maskrcnn_benchmark.modeling.matcher import Matcher 14 | from maskrcnn_benchmark.modeling.utils import cat 15 | from maskrcnn_benchmark.structures.boxlist_ops import boxlist_iou 16 | from maskrcnn_benchmark.structures.boxlist_ops import cat_boxlist 17 | from maskrcnn_benchmark.modeling.rpn.loss import RPNLossComputation 18 | 19 | class RetinaNetLossComputation(RPNLossComputation): 20 | """ 21 | This class computes the RetinaNet loss. 22 | """ 23 | 24 | def __init__(self, proposal_matcher, box_coder, 25 | generate_labels_func, 26 | sigmoid_focal_loss, 27 | bbox_reg_beta=0.11, 28 | regress_norm=1.0): 29 | """ 30 | Arguments: 31 | proposal_matcher (Matcher) 32 | box_coder (BoxCoder) 33 | """ 34 | self.proposal_matcher = proposal_matcher 35 | self.box_coder = box_coder 36 | self.box_cls_loss_func = sigmoid_focal_loss 37 | self.bbox_reg_beta = bbox_reg_beta 38 | self.copied_fields = ['labels'] 39 | self.generate_labels_func = generate_labels_func 40 | self.discard_cases = ['between_thresholds'] 41 | self.regress_norm = regress_norm 42 | 43 | def __call__(self, anchors, box_cls, box_regression, targets): 44 | """ 45 | Arguments: 46 | anchors (list[BoxList]) 47 | box_cls (list[Tensor]) 48 | box_regression (list[Tensor]) 49 | targets (list[BoxList]) 50 | 51 | Returns: 52 | retinanet_cls_loss (Tensor) 53 | retinanet_regression_loss (Tensor 54 | """ 55 | anchors = [cat_boxlist(anchors_per_image) for anchors_per_image in anchors] 56 | labels, regression_targets = self.prepare_targets(anchors, targets) 57 | 58 | N = len(labels) 59 | box_cls, box_regression = \ 60 | concat_box_prediction_layers(box_cls, box_regression) 61 | 62 | labels = torch.cat(labels, dim=0) 63 | regression_targets = torch.cat(regression_targets, dim=0) 64 | pos_inds = torch.nonzero(labels > 0).squeeze(1) 65 | 66 | retinanet_regression_loss = smooth_l1_loss( 67 | box_regression[pos_inds], 68 | regression_targets[pos_inds], 69 | beta=self.bbox_reg_beta, 70 | size_average=False, 71 | ) / (max(1, pos_inds.numel() * self.regress_norm)) 72 | 73 | labels = labels.int() 74 | 75 | retinanet_cls_loss = self.box_cls_loss_func( 76 | box_cls, 77 | labels 78 | ) / (pos_inds.numel() + N) 79 | 80 | return retinanet_cls_loss, retinanet_regression_loss 81 | 82 | 83 | def generate_retinanet_labels(matched_targets): 84 | labels_per_image = matched_targets.get_field("labels") 85 | return labels_per_image 86 | 87 | 88 | def make_retinanet_loss_evaluator(cfg, box_coder): 89 | matcher = Matcher( 90 | cfg.MODEL.RETINANET.FG_IOU_THRESHOLD, 91 | cfg.MODEL.RETINANET.BG_IOU_THRESHOLD, 92 | allow_low_quality_matches=True, 93 | ) 94 | sigmoid_focal_loss = SigmoidFocalLoss( 95 | cfg.MODEL.RETINANET.LOSS_GAMMA, 96 | cfg.MODEL.RETINANET.LOSS_ALPHA 97 | ) 98 | 99 | loss_evaluator = RetinaNetLossComputation( 100 | matcher, 101 | box_coder, 102 | generate_retinanet_labels, 103 | sigmoid_focal_loss, 104 | bbox_reg_beta = cfg.MODEL.RETINANET.BBOX_REG_BETA, 105 | regress_norm = cfg.MODEL.RETINANET.BBOX_REG_WEIGHT, 106 | ) 107 | return loss_evaluator 108 | -------------------------------------------------------------------------------- /maskrcnn_benchmark/modeling/rpn/utils.py: -------------------------------------------------------------------------------- 1 | # Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved. 2 | """ 3 | Utility functions minipulating the prediction layers 4 | """ 5 | 6 | from ..utils import cat 7 | 8 | import torch 9 | 10 | def permute_and_flatten(layer, N, A, C, H, W): 11 | layer = layer.view(N, -1, C, H, W) 12 | layer = layer.permute(0, 3, 4, 1, 2) 13 | layer = layer.reshape(N, -1, C) 14 | return layer 15 | 16 | 17 | def concat_box_prediction_layers(box_cls, box_regression): 18 | box_cls_flattened = [] 19 | box_regression_flattened = [] 20 | # for each feature level, permute the outputs to make them be in the 21 | # same format as the labels. Note that the labels are computed for 22 | # all feature levels concatenated, so we keep the same representation 23 | # for the objectness and the box_regression 24 | for box_cls_per_level, box_regression_per_level in zip( 25 | box_cls, box_regression 26 | ): 27 | N, AxC, H, W = box_cls_per_level.shape 28 | Ax4 = box_regression_per_level.shape[1] 29 | A = Ax4 // 4 30 | C = AxC // A 31 | box_cls_per_level = permute_and_flatten( 32 | box_cls_per_level, N, A, C, H, W 33 | ) 34 | box_cls_flattened.append(box_cls_per_level) 35 | 36 | box_regression_per_level = permute_and_flatten( 37 | box_regression_per_level, N, A, 4, H, W 38 | ) 39 | box_regression_flattened.append(box_regression_per_level) 40 | # concatenate on the first dimension (representing the feature levels), to 41 | # take into account the way the labels were generated (with all feature maps 42 | # being concatenated as well) 43 | box_cls = cat(box_cls_flattened, dim=1).reshape(-1, C) 44 | box_regression = cat(box_regression_flattened, dim=1).reshape(-1, 4) 45 | return box_cls, box_regression 46 | -------------------------------------------------------------------------------- /maskrcnn_benchmark/modeling/utils.py: -------------------------------------------------------------------------------- 1 | # Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved. 2 | """ 3 | Miscellaneous utility functions 4 | """ 5 | 6 | import torch 7 | 8 | 9 | def cat(tensors, dim=0): 10 | """ 11 | Efficient version of torch.cat that avoids a copy if there is only a single element in a list 12 | """ 13 | assert isinstance(tensors, (list, tuple)) 14 | if len(tensors) == 1: 15 | return tensors[0] 16 | return torch.cat(tensors, dim) 17 | -------------------------------------------------------------------------------- /maskrcnn_benchmark/solver/__init__.py: -------------------------------------------------------------------------------- 1 | # Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved. 2 | from .build import make_optimizer 3 | from .build import make_lr_scheduler 4 | from .lr_scheduler import WarmupMultiStepLR 5 | -------------------------------------------------------------------------------- /maskrcnn_benchmark/solver/build.py: -------------------------------------------------------------------------------- 1 | # Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved. 2 | import torch 3 | 4 | from .lr_scheduler import WarmupMultiStepLR 5 | 6 | 7 | def make_optimizer(cfg, model): 8 | params = [] 9 | for key, value in model.named_parameters(): 10 | if not value.requires_grad: 11 | continue 12 | lr = cfg.SOLVER.BASE_LR 13 | weight_decay = cfg.SOLVER.WEIGHT_DECAY 14 | if "bias" in key: 15 | lr = cfg.SOLVER.BASE_LR * cfg.SOLVER.BIAS_LR_FACTOR 16 | weight_decay = cfg.SOLVER.WEIGHT_DECAY_BIAS 17 | params += [{"params": [value], "lr": lr, "weight_decay": weight_decay}] 18 | 19 | optimizer = torch.optim.SGD(params, lr, momentum=cfg.SOLVER.MOMENTUM) 20 | return optimizer 21 | 22 | 23 | def make_lr_scheduler(cfg, optimizer): 24 | return WarmupMultiStepLR( 25 | optimizer, 26 | cfg.SOLVER.STEPS, 27 | cfg.SOLVER.GAMMA, 28 | warmup_factor=cfg.SOLVER.WARMUP_FACTOR, 29 | warmup_iters=cfg.SOLVER.WARMUP_ITERS, 30 | warmup_method=cfg.SOLVER.WARMUP_METHOD, 31 | ) 32 | -------------------------------------------------------------------------------- /maskrcnn_benchmark/solver/lr_scheduler.py: -------------------------------------------------------------------------------- 1 | # Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved. 2 | from bisect import bisect_right 3 | 4 | import torch 5 | 6 | 7 | # FIXME ideally this would be achieved with a CombinedLRScheduler, 8 | # separating MultiStepLR with WarmupLR 9 | # but the current LRScheduler design doesn't allow it 10 | class WarmupMultiStepLR(torch.optim.lr_scheduler._LRScheduler): 11 | def __init__( 12 | self, 13 | optimizer, 14 | milestones, 15 | gamma=0.1, 16 | warmup_factor=1.0 / 3, 17 | warmup_iters=500, 18 | warmup_method="linear", 19 | last_epoch=-1, 20 | ): 21 | if not list(milestones) == sorted(milestones): 22 | raise ValueError( 23 | "Milestones should be a list of" " increasing integers. Got {}", 24 | milestones, 25 | ) 26 | 27 | if warmup_method not in ("constant", "linear"): 28 | raise ValueError( 29 | "Only 'constant' or 'linear' warmup_method accepted" 30 | "got {}".format(warmup_method) 31 | ) 32 | self.milestones = milestones 33 | self.gamma = gamma 34 | self.warmup_factor = warmup_factor 35 | self.warmup_iters = warmup_iters 36 | self.warmup_method = warmup_method 37 | super(WarmupMultiStepLR, self).__init__(optimizer, last_epoch) 38 | 39 | def get_lr(self): 40 | warmup_factor = 1 41 | if self.last_epoch < self.warmup_iters: 42 | if self.warmup_method == "constant": 43 | warmup_factor = self.warmup_factor 44 | elif self.warmup_method == "linear": 45 | alpha = float(self.last_epoch) / self.warmup_iters 46 | warmup_factor = self.warmup_factor * (1 - alpha) + alpha 47 | return [ 48 | base_lr 49 | * warmup_factor 50 | * self.gamma ** bisect_right(self.milestones, self.last_epoch) 51 | for base_lr in self.base_lrs 52 | ] 53 | -------------------------------------------------------------------------------- /maskrcnn_benchmark/structures/__init__.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Res2Net/Res2Net-maskrcnn/dd50e0f7b767333335e18fb23054fe27e6b40519/maskrcnn_benchmark/structures/__init__.py -------------------------------------------------------------------------------- /maskrcnn_benchmark/structures/boxlist_ops.py: -------------------------------------------------------------------------------- 1 | # Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved. 2 | import torch 3 | 4 | from .bounding_box import BoxList 5 | 6 | from maskrcnn_benchmark.layers import nms as _box_nms 7 | 8 | 9 | def boxlist_nms(boxlist, nms_thresh, max_proposals=-1, score_field="scores"): 10 | """ 11 | Performs non-maximum suppression on a boxlist, with scores specified 12 | in a boxlist field via score_field. 13 | 14 | Arguments: 15 | boxlist(BoxList) 16 | nms_thresh (float) 17 | max_proposals (int): if > 0, then only the top max_proposals are kept 18 | after non-maximum suppression 19 | score_field (str) 20 | """ 21 | if nms_thresh <= 0: 22 | return boxlist 23 | mode = boxlist.mode 24 | boxlist = boxlist.convert("xyxy") 25 | boxes = boxlist.bbox 26 | score = boxlist.get_field(score_field) 27 | keep = _box_nms(boxes, score, nms_thresh) 28 | if max_proposals > 0: 29 | keep = keep[: max_proposals] 30 | boxlist = boxlist[keep] 31 | return boxlist.convert(mode) 32 | 33 | 34 | def remove_small_boxes(boxlist, min_size): 35 | """ 36 | Only keep boxes with both sides >= min_size 37 | 38 | Arguments: 39 | boxlist (Boxlist) 40 | min_size (int) 41 | """ 42 | # TODO maybe add an API for querying the ws / hs 43 | xywh_boxes = boxlist.convert("xywh").bbox 44 | _, _, ws, hs = xywh_boxes.unbind(dim=1) 45 | keep = ( 46 | (ws >= min_size) & (hs >= min_size) 47 | ).nonzero().squeeze(1) 48 | return boxlist[keep] 49 | 50 | 51 | # implementation from https://github.com/kuangliu/torchcv/blob/master/torchcv/utils/box.py 52 | # with slight modifications 53 | def boxlist_iou(boxlist1, boxlist2): 54 | """Compute the intersection over union of two set of boxes. 55 | The box order must be (xmin, ymin, xmax, ymax). 56 | 57 | Arguments: 58 | box1: (BoxList) bounding boxes, sized [N,4]. 59 | box2: (BoxList) bounding boxes, sized [M,4]. 60 | 61 | Returns: 62 | (tensor) iou, sized [N,M]. 63 | 64 | Reference: 65 | https://github.com/chainer/chainercv/blob/master/chainercv/utils/bbox/bbox_iou.py 66 | """ 67 | if boxlist1.size != boxlist2.size: 68 | raise RuntimeError( 69 | "boxlists should have same image size, got {}, {}".format(boxlist1, boxlist2)) 70 | 71 | N = len(boxlist1) 72 | M = len(boxlist2) 73 | 74 | area1 = boxlist1.area() 75 | area2 = boxlist2.area() 76 | 77 | box1, box2 = boxlist1.bbox, boxlist2.bbox 78 | 79 | lt = torch.max(box1[:, None, :2], box2[:, :2]) # [N,M,2] 80 | rb = torch.min(box1[:, None, 2:], box2[:, 2:]) # [N,M,2] 81 | 82 | TO_REMOVE = 1 83 | 84 | wh = (rb - lt + TO_REMOVE).clamp(min=0) # [N,M,2] 85 | inter = wh[:, :, 0] * wh[:, :, 1] # [N,M] 86 | 87 | iou = inter / (area1[:, None] + area2 - inter) 88 | return iou 89 | 90 | 91 | # TODO redundant, remove 92 | def _cat(tensors, dim=0): 93 | """ 94 | Efficient version of torch.cat that avoids a copy if there is only a single element in a list 95 | """ 96 | assert isinstance(tensors, (list, tuple)) 97 | if len(tensors) == 1: 98 | return tensors[0] 99 | return torch.cat(tensors, dim) 100 | 101 | 102 | def cat_boxlist(bboxes): 103 | """ 104 | Concatenates a list of BoxList (having the same image size) into a 105 | single BoxList 106 | 107 | Arguments: 108 | bboxes (list[BoxList]) 109 | """ 110 | assert isinstance(bboxes, (list, tuple)) 111 | assert all(isinstance(bbox, BoxList) for bbox in bboxes) 112 | 113 | size = bboxes[0].size 114 | assert all(bbox.size == size for bbox in bboxes) 115 | 116 | mode = bboxes[0].mode 117 | assert all(bbox.mode == mode for bbox in bboxes) 118 | 119 | fields = set(bboxes[0].fields()) 120 | assert all(set(bbox.fields()) == fields for bbox in bboxes) 121 | 122 | cat_boxes = BoxList(_cat([bbox.bbox for bbox in bboxes], dim=0), size, mode) 123 | 124 | for field in fields: 125 | data = _cat([bbox.get_field(field) for bbox in bboxes], dim=0) 126 | cat_boxes.add_field(field, data) 127 | 128 | return cat_boxes 129 | -------------------------------------------------------------------------------- /maskrcnn_benchmark/structures/image_list.py: -------------------------------------------------------------------------------- 1 | # Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved. 2 | from __future__ import division 3 | 4 | import torch 5 | 6 | 7 | class ImageList(object): 8 | """ 9 | Structure that holds a list of images (of possibly 10 | varying sizes) as a single tensor. 11 | This works by padding the images to the same size, 12 | and storing in a field the original sizes of each image 13 | """ 14 | 15 | def __init__(self, tensors, image_sizes): 16 | """ 17 | Arguments: 18 | tensors (tensor) 19 | image_sizes (list[tuple[int, int]]) 20 | """ 21 | self.tensors = tensors 22 | self.image_sizes = image_sizes 23 | 24 | def to(self, *args, **kwargs): 25 | cast_tensor = self.tensors.to(*args, **kwargs) 26 | return ImageList(cast_tensor, self.image_sizes) 27 | 28 | 29 | def to_image_list(tensors, size_divisible=0): 30 | """ 31 | tensors can be an ImageList, a torch.Tensor or 32 | an iterable of Tensors. It can't be a numpy array. 33 | When tensors is an iterable of Tensors, it pads 34 | the Tensors with zeros so that they have the same 35 | shape 36 | """ 37 | if isinstance(tensors, torch.Tensor) and size_divisible > 0: 38 | tensors = [tensors] 39 | 40 | if isinstance(tensors, ImageList): 41 | return tensors 42 | elif isinstance(tensors, torch.Tensor): 43 | # single tensor shape can be inferred 44 | assert tensors.dim() == 4 45 | image_sizes = [tensor.shape[-2:] for tensor in tensors] 46 | return ImageList(tensors, image_sizes) 47 | elif isinstance(tensors, (tuple, list)): 48 | max_size = tuple(max(s) for s in zip(*[img.shape for img in tensors])) 49 | 50 | # TODO Ideally, just remove this and let me model handle arbitrary 51 | # input sizs 52 | if size_divisible > 0: 53 | import math 54 | 55 | stride = size_divisible 56 | max_size = list(max_size) 57 | max_size[1] = int(math.ceil(max_size[1] / stride) * stride) 58 | max_size[2] = int(math.ceil(max_size[2] / stride) * stride) 59 | max_size = tuple(max_size) 60 | 61 | batch_shape = (len(tensors),) + max_size 62 | batched_imgs = tensors[0].new(*batch_shape).zero_() 63 | for img, pad_img in zip(tensors, batched_imgs): 64 | pad_img[: img.shape[0], : img.shape[1], : img.shape[2]].copy_(img) 65 | 66 | image_sizes = [im.shape[-2:] for im in tensors] 67 | 68 | return ImageList(batched_imgs, image_sizes) 69 | else: 70 | raise TypeError("Unsupported type for to_image_list: {}".format(type(tensors))) 71 | -------------------------------------------------------------------------------- /maskrcnn_benchmark/utils/README.md: -------------------------------------------------------------------------------- 1 | # Utility functions 2 | 3 | This folder contain utility functions that are not used in the 4 | core library, but are useful for building models or training 5 | code using the config system. 6 | -------------------------------------------------------------------------------- /maskrcnn_benchmark/utils/__init__.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Res2Net/Res2Net-maskrcnn/dd50e0f7b767333335e18fb23054fe27e6b40519/maskrcnn_benchmark/utils/__init__.py -------------------------------------------------------------------------------- /maskrcnn_benchmark/utils/collect_env.py: -------------------------------------------------------------------------------- 1 | # Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved. 2 | import PIL 3 | 4 | from torch.utils.collect_env import get_pretty_env_info 5 | 6 | 7 | def get_pil_version(): 8 | return "\n Pillow ({})".format(PIL.__version__) 9 | 10 | 11 | def collect_env_info(): 12 | env_str = get_pretty_env_info() 13 | env_str += get_pil_version() 14 | return env_str 15 | -------------------------------------------------------------------------------- /maskrcnn_benchmark/utils/comm.py: -------------------------------------------------------------------------------- 1 | """ 2 | This file contains primitives for multi-gpu communication. 3 | This is useful when doing distributed training. 4 | """ 5 | 6 | import pickle 7 | import time 8 | 9 | import torch 10 | import torch.distributed as dist 11 | 12 | 13 | def get_world_size(): 14 | if not dist.is_available(): 15 | return 1 16 | if not dist.is_initialized(): 17 | return 1 18 | return dist.get_world_size() 19 | 20 | 21 | def get_rank(): 22 | if not dist.is_available(): 23 | return 0 24 | if not dist.is_initialized(): 25 | return 0 26 | return dist.get_rank() 27 | 28 | 29 | def is_main_process(): 30 | return get_rank() == 0 31 | 32 | 33 | def synchronize(): 34 | """ 35 | Helper function to synchronize (barrier) among all processes when 36 | using distributed training 37 | """ 38 | if not dist.is_available(): 39 | return 40 | if not dist.is_initialized(): 41 | return 42 | world_size = dist.get_world_size() 43 | if world_size == 1: 44 | return 45 | dist.barrier() 46 | 47 | 48 | def all_gather(data): 49 | """ 50 | Run all_gather on arbitrary picklable data (not necessarily tensors) 51 | Args: 52 | data: any picklable object 53 | Returns: 54 | list[data]: list of data gathered from each rank 55 | """ 56 | world_size = get_world_size() 57 | if world_size == 1: 58 | return [data] 59 | 60 | # serialized to a Tensor 61 | buffer = pickle.dumps(data) 62 | storage = torch.ByteStorage.from_buffer(buffer) 63 | tensor = torch.ByteTensor(storage).to("cuda") 64 | 65 | # obtain Tensor size of each rank 66 | local_size = torch.IntTensor([tensor.numel()]).to("cuda") 67 | size_list = [torch.IntTensor([0]).to("cuda") for _ in range(world_size)] 68 | dist.all_gather(size_list, local_size) 69 | size_list = [int(size.item()) for size in size_list] 70 | max_size = max(size_list) 71 | 72 | # receiving Tensor from all ranks 73 | # we pad the tensor because torch all_gather does not support 74 | # gathering tensors of different shapes 75 | tensor_list = [] 76 | for _ in size_list: 77 | tensor_list.append(torch.ByteTensor(size=(max_size,)).to("cuda")) 78 | if local_size != max_size: 79 | padding = torch.ByteTensor(size=(max_size - local_size,)).to("cuda") 80 | tensor = torch.cat((tensor, padding), dim=0) 81 | dist.all_gather(tensor_list, tensor) 82 | 83 | data_list = [] 84 | for size, tensor in zip(size_list, tensor_list): 85 | buffer = tensor.cpu().numpy().tobytes()[:size] 86 | data_list.append(pickle.loads(buffer)) 87 | 88 | return data_list 89 | 90 | 91 | def reduce_dict(input_dict, average=True): 92 | """ 93 | Args: 94 | input_dict (dict): all the values will be reduced 95 | average (bool): whether to do average or sum 96 | Reduce the values in the dictionary from all processes so that process with rank 97 | 0 has the averaged results. Returns a dict with the same fields as 98 | input_dict, after reduction. 99 | """ 100 | world_size = get_world_size() 101 | if world_size < 2: 102 | return input_dict 103 | with torch.no_grad(): 104 | names = [] 105 | values = [] 106 | # sort the keys so that they are consistent across processes 107 | for k in sorted(input_dict.keys()): 108 | names.append(k) 109 | values.append(input_dict[k]) 110 | values = torch.stack(values, dim=0) 111 | dist.reduce(values, dst=0) 112 | if dist.get_rank() == 0 and average: 113 | # only main process gets accumulated, so only divide by 114 | # world_size in this case 115 | values /= world_size 116 | reduced_dict = {k: v for k, v in zip(names, values)} 117 | return reduced_dict 118 | -------------------------------------------------------------------------------- /maskrcnn_benchmark/utils/cv2_util.py: -------------------------------------------------------------------------------- 1 | """ 2 | Module for cv2 utility functions and maintaining version compatibility 3 | between 3.x and 4.x 4 | """ 5 | import cv2 6 | 7 | 8 | def findContours(*args, **kwargs): 9 | """ 10 | Wraps cv2.findContours to maintain compatiblity between versions 11 | 3 and 4 12 | 13 | Returns: 14 | contours, hierarchy 15 | """ 16 | if cv2.__version__.startswith('4'): 17 | contours, hierarchy = cv2.findContours(*args, **kwargs) 18 | elif cv2.__version__.startswith('3'): 19 | _, contours, hierarchy = cv2.findContours(*args, **kwargs) 20 | else: 21 | raise AssertionError( 22 | 'cv2 must be either version 3 or 4 to call this method') 23 | 24 | return contours, hierarchy 25 | -------------------------------------------------------------------------------- /maskrcnn_benchmark/utils/env.py: -------------------------------------------------------------------------------- 1 | # Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved. 2 | import os 3 | 4 | from maskrcnn_benchmark.utils.imports import import_file 5 | 6 | 7 | def setup_environment(): 8 | """Perform environment setup work. The default setup is a no-op, but this 9 | function allows the user to specify a Python source file that performs 10 | custom setup work that may be necessary to their computing environment. 11 | """ 12 | custom_module_path = os.environ.get("TORCH_DETECTRON_ENV_MODULE") 13 | if custom_module_path: 14 | setup_custom_environment(custom_module_path) 15 | else: 16 | # The default setup is a no-op 17 | pass 18 | 19 | 20 | def setup_custom_environment(custom_module_path): 21 | """Load custom environment setup from a Python source file and run the setup 22 | function. 23 | """ 24 | module = import_file("maskrcnn_benchmark.utils.env.custom_module", custom_module_path) 25 | assert hasattr(module, "setup_environment") and callable( 26 | module.setup_environment 27 | ), ( 28 | "Custom environment module defined in {} does not have the " 29 | "required callable attribute 'setup_environment'." 30 | ).format( 31 | custom_module_path 32 | ) 33 | module.setup_environment() 34 | 35 | 36 | # Force environment setup when this module is imported 37 | setup_environment() 38 | -------------------------------------------------------------------------------- /maskrcnn_benchmark/utils/imports.py: -------------------------------------------------------------------------------- 1 | # Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved. 2 | import torch 3 | 4 | if torch._six.PY3: 5 | import importlib 6 | import importlib.util 7 | import sys 8 | 9 | 10 | # from https://stackoverflow.com/questions/67631/how-to-import-a-module-given-the-full-path?utm_medium=organic&utm_source=google_rich_qa&utm_campaign=google_rich_qa 11 | def import_file(module_name, file_path, make_importable=False): 12 | spec = importlib.util.spec_from_file_location(module_name, file_path) 13 | module = importlib.util.module_from_spec(spec) 14 | spec.loader.exec_module(module) 15 | if make_importable: 16 | sys.modules[module_name] = module 17 | return module 18 | else: 19 | import imp 20 | 21 | def import_file(module_name, file_path, make_importable=None): 22 | module = imp.load_source(module_name, file_path) 23 | return module 24 | -------------------------------------------------------------------------------- /maskrcnn_benchmark/utils/logger.py: -------------------------------------------------------------------------------- 1 | # Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved. 2 | import logging 3 | import os 4 | import sys 5 | 6 | 7 | def setup_logger(name, save_dir, distributed_rank, filename="log.txt"): 8 | logger = logging.getLogger(name) 9 | logger.setLevel(logging.DEBUG) 10 | # don't log results for the non-master process 11 | if distributed_rank > 0: 12 | return logger 13 | ch = logging.StreamHandler(stream=sys.stdout) 14 | ch.setLevel(logging.DEBUG) 15 | formatter = logging.Formatter("%(asctime)s %(name)s %(levelname)s: %(message)s") 16 | ch.setFormatter(formatter) 17 | logger.addHandler(ch) 18 | 19 | if save_dir: 20 | fh = logging.FileHandler(os.path.join(save_dir, filename)) 21 | fh.setLevel(logging.DEBUG) 22 | fh.setFormatter(formatter) 23 | logger.addHandler(fh) 24 | 25 | return logger 26 | -------------------------------------------------------------------------------- /maskrcnn_benchmark/utils/metric_logger.py: -------------------------------------------------------------------------------- 1 | # Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved. 2 | from collections import defaultdict 3 | from collections import deque 4 | 5 | import torch 6 | 7 | 8 | class SmoothedValue(object): 9 | """Track a series of values and provide access to smoothed values over a 10 | window or the global series average. 11 | """ 12 | 13 | def __init__(self, window_size=20): 14 | self.deque = deque(maxlen=window_size) 15 | self.series = [] 16 | self.total = 0.0 17 | self.count = 0 18 | 19 | def update(self, value): 20 | self.deque.append(value) 21 | self.series.append(value) 22 | self.count += 1 23 | self.total += value 24 | 25 | @property 26 | def median(self): 27 | d = torch.tensor(list(self.deque)) 28 | return d.median().item() 29 | 30 | @property 31 | def avg(self): 32 | d = torch.tensor(list(self.deque)) 33 | return d.mean().item() 34 | 35 | @property 36 | def global_avg(self): 37 | return self.total / self.count 38 | 39 | 40 | class MetricLogger(object): 41 | def __init__(self, delimiter="\t"): 42 | self.meters = defaultdict(SmoothedValue) 43 | self.delimiter = delimiter 44 | 45 | def update(self, **kwargs): 46 | for k, v in kwargs.items(): 47 | if isinstance(v, torch.Tensor): 48 | v = v.item() 49 | assert isinstance(v, (float, int)) 50 | self.meters[k].update(v) 51 | 52 | def __getattr__(self, attr): 53 | if attr in self.meters: 54 | return self.meters[attr] 55 | if attr in self.__dict__: 56 | return self.__dict__[attr] 57 | raise AttributeError("'{}' object has no attribute '{}'".format( 58 | type(self).__name__, attr)) 59 | 60 | def __str__(self): 61 | loss_str = [] 62 | for name, meter in self.meters.items(): 63 | loss_str.append( 64 | "{}: {:.4f} ({:.4f})".format(name, meter.median, meter.global_avg) 65 | ) 66 | return self.delimiter.join(loss_str) 67 | -------------------------------------------------------------------------------- /maskrcnn_benchmark/utils/miscellaneous.py: -------------------------------------------------------------------------------- 1 | # Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved. 2 | import errno 3 | import os 4 | 5 | 6 | def mkdir(path): 7 | try: 8 | os.makedirs(path) 9 | except OSError as e: 10 | if e.errno != errno.EEXIST: 11 | raise 12 | -------------------------------------------------------------------------------- /maskrcnn_benchmark/utils/model_serialization.py: -------------------------------------------------------------------------------- 1 | # Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved. 2 | from collections import OrderedDict 3 | import logging 4 | 5 | import torch 6 | 7 | from maskrcnn_benchmark.utils.imports import import_file 8 | 9 | 10 | def align_and_update_state_dicts(model_state_dict, loaded_state_dict): 11 | """ 12 | Strategy: suppose that the models that we will create will have prefixes appended 13 | to each of its keys, for example due to an extra level of nesting that the original 14 | pre-trained weights from ImageNet won't contain. For example, model.state_dict() 15 | might return backbone[0].body.res2.conv1.weight, while the pre-trained model contains 16 | res2.conv1.weight. We thus want to match both parameters together. 17 | For that, we look for each model weight, look among all loaded keys if there is one 18 | that is a suffix of the current weight name, and use it if that's the case. 19 | If multiple matches exist, take the one with longest size 20 | of the corresponding name. For example, for the same model as before, the pretrained 21 | weight file can contain both res2.conv1.weight, as well as conv1.weight. In this case, 22 | we want to match backbone[0].body.conv1.weight to conv1.weight, and 23 | backbone[0].body.res2.conv1.weight to res2.conv1.weight. 24 | """ 25 | current_keys = sorted(list(model_state_dict.keys())) 26 | loaded_keys = sorted(list(loaded_state_dict.keys())) 27 | # get a matrix of string matches, where each (i, j) entry correspond to the size of the 28 | # loaded_key string, if it matches 29 | match_matrix = [ 30 | len(j) if i.endswith(j) else 0 for i in current_keys for j in loaded_keys 31 | ] 32 | match_matrix = torch.as_tensor(match_matrix).view( 33 | len(current_keys), len(loaded_keys) 34 | ) 35 | max_match_size, idxs = match_matrix.max(1) 36 | # remove indices that correspond to no-match 37 | idxs[max_match_size == 0] = -1 38 | 39 | # used for logging 40 | max_size = max([len(key) for key in current_keys]) if current_keys else 1 41 | max_size_loaded = max([len(key) for key in loaded_keys]) if loaded_keys else 1 42 | log_str_template = "{: <{}} loaded from {: <{}} of shape {}" 43 | logger = logging.getLogger(__name__) 44 | for idx_new, idx_old in enumerate(idxs.tolist()): 45 | if idx_old == -1: 46 | continue 47 | key = current_keys[idx_new] 48 | key_old = loaded_keys[idx_old] 49 | model_state_dict[key] = loaded_state_dict[key_old] 50 | logger.info( 51 | log_str_template.format( 52 | key, 53 | max_size, 54 | key_old, 55 | max_size_loaded, 56 | tuple(loaded_state_dict[key_old].shape), 57 | ) 58 | ) 59 | 60 | 61 | def strip_prefix_if_present(state_dict, prefix): 62 | keys = sorted(state_dict.keys()) 63 | if not all(key.startswith(prefix) for key in keys): 64 | return state_dict 65 | stripped_state_dict = OrderedDict() 66 | for key, value in state_dict.items(): 67 | stripped_state_dict[key.replace(prefix, "")] = value 68 | return stripped_state_dict 69 | 70 | 71 | def load_state_dict(model, loaded_state_dict): 72 | model_state_dict = model.state_dict() 73 | # if the state_dict comes from a model that was wrapped in a 74 | # DataParallel or DistributedDataParallel during serialization, 75 | # remove the "module" prefix before performing the matching 76 | loaded_state_dict = strip_prefix_if_present(loaded_state_dict, prefix="module.") 77 | align_and_update_state_dicts(model_state_dict, loaded_state_dict) 78 | 79 | # use strict loading 80 | model.load_state_dict(model_state_dict) 81 | -------------------------------------------------------------------------------- /maskrcnn_benchmark/utils/model_zoo.py: -------------------------------------------------------------------------------- 1 | # Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved. 2 | import os 3 | import sys 4 | 5 | from torch.utils.model_zoo import _download_url_to_file 6 | from torch.utils.model_zoo import urlparse 7 | from torch.utils.model_zoo import HASH_REGEX 8 | 9 | from maskrcnn_benchmark.utils.comm import is_main_process 10 | from maskrcnn_benchmark.utils.comm import synchronize 11 | 12 | 13 | # very similar to https://github.com/pytorch/pytorch/blob/master/torch/utils/model_zoo.py 14 | # but with a few improvements and modifications 15 | def cache_url(url, model_dir=None, progress=True): 16 | r"""Loads the Torch serialized object at the given URL. 17 | If the object is already present in `model_dir`, it's deserialized and 18 | returned. The filename part of the URL should follow the naming convention 19 | ``filename-.ext`` where ```` is the first eight or more 20 | digits of the SHA256 hash of the contents of the file. The hash is used to 21 | ensure unique names and to verify the contents of the file. 22 | The default value of `model_dir` is ``$TORCH_HOME/models`` where 23 | ``$TORCH_HOME`` defaults to ``~/.torch``. The default directory can be 24 | overridden with the ``$TORCH_MODEL_ZOO`` environment variable. 25 | Args: 26 | url (string): URL of the object to download 27 | model_dir (string, optional): directory in which to save the object 28 | progress (bool, optional): whether or not to display a progress bar to stderr 29 | Example: 30 | >>> cached_file = maskrcnn_benchmark.utils.model_zoo.cache_url('https://s3.amazonaws.com/pytorch/models/resnet18-5c106cde.pth') 31 | """ 32 | if model_dir is None: 33 | torch_home = os.path.expanduser(os.getenv('TORCH_HOME', '~/.torch')) 34 | model_dir = os.getenv('TORCH_MODEL_ZOO', os.path.join(torch_home, 'models')) 35 | if not os.path.exists(model_dir): 36 | os.makedirs(model_dir) 37 | parts = urlparse(url) 38 | filename = os.path.basename(parts.path) 39 | if filename == "model_final.pkl": 40 | # workaround as pre-trained Caffe2 models from Detectron have all the same filename 41 | # so make the full path the filename by replacing / with _ 42 | filename = parts.path.replace("/", "_") 43 | cached_file = os.path.join(model_dir, filename) 44 | if not os.path.exists(cached_file) and is_main_process(): 45 | sys.stderr.write('Downloading: "{}" to {}\n'.format(url, cached_file)) 46 | hash_prefix = HASH_REGEX.search(filename) 47 | if hash_prefix is not None: 48 | hash_prefix = hash_prefix.group(1) 49 | # workaround: Caffe2 models don't have a hash, but follow the R-50 convention, 50 | # which matches the hash PyTorch uses. So we skip the hash matching 51 | # if the hash_prefix is less than 6 characters 52 | if len(hash_prefix) < 6: 53 | hash_prefix = None 54 | _download_url_to_file(url, cached_file, hash_prefix, progress=progress) 55 | synchronize() 56 | return cached_file 57 | -------------------------------------------------------------------------------- /maskrcnn_benchmark/utils/registry.py: -------------------------------------------------------------------------------- 1 | # Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved. 2 | 3 | 4 | def _register_generic(module_dict, module_name, module): 5 | assert module_name not in module_dict 6 | module_dict[module_name] = module 7 | 8 | 9 | class Registry(dict): 10 | ''' 11 | A helper class for managing registering modules, it extends a dictionary 12 | and provides a register functions. 13 | 14 | Eg. creeting a registry: 15 | some_registry = Registry({"default": default_module}) 16 | 17 | There're two ways of registering new modules: 18 | 1): normal way is just calling register function: 19 | def foo(): 20 | ... 21 | some_registry.register("foo_module", foo) 22 | 2): used as decorator when declaring the module: 23 | @some_registry.register("foo_module") 24 | @some_registry.register("foo_modeul_nickname") 25 | def foo(): 26 | ... 27 | 28 | Access of module is just like using a dictionary, eg: 29 | f = some_registry["foo_modeul"] 30 | ''' 31 | def __init__(self, *args, **kwargs): 32 | super(Registry, self).__init__(*args, **kwargs) 33 | 34 | def register(self, module_name, module=None): 35 | # used as function call 36 | if module is not None: 37 | _register_generic(self, module_name, module) 38 | return 39 | 40 | # used as decorator 41 | def register_fn(fn): 42 | _register_generic(self, module_name, fn) 43 | return fn 44 | 45 | return register_fn 46 | -------------------------------------------------------------------------------- /setup.py: -------------------------------------------------------------------------------- 1 | # Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved. 2 | #!/usr/bin/env python 3 | 4 | import glob 5 | import os 6 | 7 | import torch 8 | from setuptools import find_packages 9 | from setuptools import setup 10 | from torch.utils.cpp_extension import CUDA_HOME 11 | from torch.utils.cpp_extension import CppExtension 12 | from torch.utils.cpp_extension import CUDAExtension 13 | 14 | requirements = ["torch", "torchvision"] 15 | 16 | 17 | def get_extensions(): 18 | this_dir = os.path.dirname(os.path.abspath(__file__)) 19 | extensions_dir = os.path.join(this_dir, "maskrcnn_benchmark", "csrc") 20 | 21 | main_file = glob.glob(os.path.join(extensions_dir, "*.cpp")) 22 | source_cpu = glob.glob(os.path.join(extensions_dir, "cpu", "*.cpp")) 23 | source_cuda = glob.glob(os.path.join(extensions_dir, "cuda", "*.cu")) 24 | 25 | sources = main_file + source_cpu 26 | extension = CppExtension 27 | 28 | extra_compile_args = {"cxx": []} 29 | define_macros = [] 30 | 31 | if torch.cuda.is_available() and CUDA_HOME is not None: 32 | extension = CUDAExtension 33 | sources += source_cuda 34 | define_macros += [("WITH_CUDA", None)] 35 | extra_compile_args["nvcc"] = [ 36 | "-DCUDA_HAS_FP16=1", 37 | "-D__CUDA_NO_HALF_OPERATORS__", 38 | "-D__CUDA_NO_HALF_CONVERSIONS__", 39 | "-D__CUDA_NO_HALF2_OPERATORS__", 40 | ] 41 | 42 | sources = [os.path.join(extensions_dir, s) for s in sources] 43 | 44 | include_dirs = [extensions_dir] 45 | 46 | ext_modules = [ 47 | extension( 48 | "maskrcnn_benchmark._C", 49 | sources, 50 | include_dirs=include_dirs, 51 | define_macros=define_macros, 52 | extra_compile_args=extra_compile_args, 53 | ) 54 | ] 55 | 56 | return ext_modules 57 | 58 | 59 | setup( 60 | name="maskrcnn_benchmark", 61 | version="0.1", 62 | author="fmassa", 63 | url="https://github.com/facebookresearch/maskrcnn-benchmark", 64 | description="object detection in pytorch", 65 | packages=find_packages(exclude=("configs", "tests",)), 66 | # install_requires=requirements, 67 | ext_modules=get_extensions(), 68 | cmdclass={"build_ext": torch.utils.cpp_extension.BuildExtension}, 69 | ) 70 | -------------------------------------------------------------------------------- /tests/env_tests/env.py: -------------------------------------------------------------------------------- 1 | # Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved. 2 | 3 | import os 4 | 5 | 6 | def get_config_root_path(): 7 | ''' Path to configs for unit tests ''' 8 | # cur_file_dir is root/tests/env_tests 9 | cur_file_dir = os.path.dirname(os.path.abspath(os.path.realpath(__file__))) 10 | ret = os.path.dirname(os.path.dirname(cur_file_dir)) 11 | ret = os.path.join(ret, "configs") 12 | return ret 13 | -------------------------------------------------------------------------------- /tests/test_backbones.py: -------------------------------------------------------------------------------- 1 | # Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved. 2 | 3 | import unittest 4 | import copy 5 | import torch 6 | # import modules to to register backbones 7 | from maskrcnn_benchmark.modeling.backbone import build_backbone # NoQA 8 | from maskrcnn_benchmark.modeling import registry 9 | from maskrcnn_benchmark.config import cfg as g_cfg 10 | from utils import load_config 11 | 12 | 13 | # overwrite configs if specified, otherwise default config is used 14 | BACKBONE_CFGS = { 15 | "R-50-FPN": "e2e_faster_rcnn_R_50_FPN_1x.yaml", 16 | "R-101-FPN": "e2e_faster_rcnn_R_101_FPN_1x.yaml", 17 | "R-152-FPN": "e2e_faster_rcnn_R_101_FPN_1x.yaml", 18 | "R-50-FPN-RETINANET": "retinanet/retinanet_R-50-FPN_1x.yaml", 19 | "R-101-FPN-RETINANET": "retinanet/retinanet_R-101-FPN_1x.yaml", 20 | } 21 | 22 | 23 | class TestBackbones(unittest.TestCase): 24 | def test_build_backbones(self): 25 | ''' Make sure backbones run ''' 26 | 27 | self.assertGreater(len(registry.BACKBONES), 0) 28 | 29 | for name, backbone_builder in registry.BACKBONES.items(): 30 | print('Testing {}...'.format(name)) 31 | if name in BACKBONE_CFGS: 32 | cfg = load_config(BACKBONE_CFGS[name]) 33 | else: 34 | # Use default config if config file is not specified 35 | cfg = copy.deepcopy(g_cfg) 36 | backbone = backbone_builder(cfg) 37 | 38 | # make sures the backbone has `out_channels` 39 | self.assertIsNotNone( 40 | getattr(backbone, 'out_channels', None), 41 | 'Need to provide out_channels for backbone {}'.format(name) 42 | ) 43 | 44 | N, C_in, H, W = 2, 3, 224, 256 45 | input = torch.rand([N, C_in, H, W], dtype=torch.float32) 46 | out = backbone(input) 47 | for cur_out in out: 48 | self.assertEqual( 49 | cur_out.shape[:2], 50 | torch.Size([N, backbone.out_channels]) 51 | ) 52 | 53 | 54 | if __name__ == "__main__": 55 | unittest.main() 56 | -------------------------------------------------------------------------------- /tests/test_box_coder.py: -------------------------------------------------------------------------------- 1 | # Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved. 2 | 3 | import unittest 4 | 5 | import numpy as np 6 | import torch 7 | from maskrcnn_benchmark.modeling.box_coder import BoxCoder 8 | 9 | 10 | class TestBoxCoder(unittest.TestCase): 11 | def test_box_decoder(self): 12 | """ Match unit test UtilsBoxesTest.TestBboxTransformRandom in 13 | caffe2/operators/generate_proposals_op_util_boxes_test.cc 14 | """ 15 | box_coder = BoxCoder(weights=(1.0, 1.0, 1.0, 1.0)) 16 | bbox = torch.from_numpy( 17 | np.array( 18 | [ 19 | 175.62031555, 20 | 20.91103172, 21 | 253.352005, 22 | 155.0145874, 23 | 169.24636841, 24 | 4.85241556, 25 | 228.8605957, 26 | 105.02092743, 27 | 181.77426147, 28 | 199.82876587, 29 | 192.88427734, 30 | 214.0255127, 31 | 174.36262512, 32 | 186.75761414, 33 | 296.19091797, 34 | 231.27906799, 35 | 22.73153877, 36 | 92.02596283, 37 | 135.5695343, 38 | 208.80291748, 39 | ] 40 | ) 41 | .astype(np.float32) 42 | .reshape(-1, 4) 43 | ) 44 | 45 | deltas = torch.from_numpy( 46 | np.array( 47 | [ 48 | 0.47861834, 49 | 0.13992102, 50 | 0.14961673, 51 | 0.71495209, 52 | 0.29915856, 53 | -0.35664671, 54 | 0.89018666, 55 | 0.70815367, 56 | -0.03852064, 57 | 0.44466892, 58 | 0.49492538, 59 | 0.71409376, 60 | 0.28052918, 61 | 0.02184832, 62 | 0.65289006, 63 | 1.05060139, 64 | -0.38172557, 65 | -0.08533806, 66 | -0.60335309, 67 | 0.79052375, 68 | ] 69 | ) 70 | .astype(np.float32) 71 | .reshape(-1, 4) 72 | ) 73 | 74 | gt_bbox = ( 75 | np.array( 76 | [ 77 | 206.949539, 78 | -30.715202, 79 | 297.387665, 80 | 244.448486, 81 | 143.871216, 82 | -83.342888, 83 | 290.502289, 84 | 121.053398, 85 | 177.430283, 86 | 198.666245, 87 | 196.295273, 88 | 228.703079, 89 | 152.251892, 90 | 145.431564, 91 | 387.215454, 92 | 274.594238, 93 | 5.062420, 94 | 11.040955, 95 | 66.328903, 96 | 269.686218, 97 | ] 98 | ) 99 | .astype(np.float32) 100 | .reshape(-1, 4) 101 | ) 102 | 103 | results = box_coder.decode(deltas, bbox) 104 | 105 | np.testing.assert_allclose(results.detach().numpy(), gt_bbox, atol=1e-4) 106 | 107 | 108 | if __name__ == "__main__": 109 | unittest.main() 110 | -------------------------------------------------------------------------------- /tests/test_configs.py: -------------------------------------------------------------------------------- 1 | # Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved. 2 | 3 | import unittest 4 | import glob 5 | import os 6 | import utils 7 | 8 | 9 | class TestConfigs(unittest.TestCase): 10 | def test_configs_load(self): 11 | ''' Make sure configs are loadable ''' 12 | 13 | cfg_root_path = utils.get_config_root_path() 14 | files = glob.glob( 15 | os.path.join(cfg_root_path, "./**/*.yaml"), recursive=True) 16 | self.assertGreater(len(files), 0) 17 | 18 | for fn in files: 19 | print('Loading {}...'.format(fn)) 20 | utils.load_config_from_file(fn) 21 | 22 | 23 | if __name__ == "__main__": 24 | unittest.main() 25 | -------------------------------------------------------------------------------- /tests/test_fbnet.py: -------------------------------------------------------------------------------- 1 | # Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved. 2 | 3 | import unittest 4 | 5 | import numpy as np 6 | import torch 7 | import maskrcnn_benchmark.modeling.backbone.fbnet_builder as fbnet_builder 8 | 9 | 10 | TEST_CUDA = torch.cuda.is_available() 11 | 12 | 13 | def _test_primitive(self, device, op_name, op_func, N, C_in, C_out, expand, stride): 14 | op = op_func(C_in, C_out, expand, stride).to(device) 15 | input = torch.rand([N, C_in, 7, 7], dtype=torch.float32).to(device) 16 | output = op(input) 17 | self.assertEqual( 18 | output.shape[:2], torch.Size([N, C_out]), 19 | 'Primitive {} failed for shape {}.'.format(op_name, input.shape) 20 | ) 21 | 22 | 23 | class TestFBNetBuilder(unittest.TestCase): 24 | def test_identity(self): 25 | id_op = fbnet_builder.Identity(20, 20, 1) 26 | input = torch.rand([10, 20, 7, 7], dtype=torch.float32) 27 | output = id_op(input) 28 | np.testing.assert_array_equal(np.array(input), np.array(output)) 29 | 30 | id_op = fbnet_builder.Identity(20, 40, 2) 31 | input = torch.rand([10, 20, 7, 7], dtype=torch.float32) 32 | output = id_op(input) 33 | np.testing.assert_array_equal(output.shape, [10, 40, 4, 4]) 34 | 35 | def test_primitives(self): 36 | ''' Make sures the primitives runs ''' 37 | for op_name, op_func in fbnet_builder.PRIMITIVES.items(): 38 | print('Testing {}'.format(op_name)) 39 | 40 | _test_primitive( 41 | self, "cpu", 42 | op_name, op_func, 43 | N=20, C_in=16, C_out=32, expand=4, stride=1 44 | ) 45 | 46 | @unittest.skipIf(not TEST_CUDA, "no CUDA detected") 47 | def test_primitives_cuda(self): 48 | ''' Make sures the primitives runs on cuda ''' 49 | for op_name, op_func in fbnet_builder.PRIMITIVES.items(): 50 | print('Testing {}'.format(op_name)) 51 | 52 | _test_primitive( 53 | self, "cuda", 54 | op_name, op_func, 55 | N=20, C_in=16, C_out=32, expand=4, stride=1 56 | ) 57 | 58 | def test_primitives_empty_batch(self): 59 | ''' Make sures the primitives runs ''' 60 | for op_name, op_func in fbnet_builder.PRIMITIVES.items(): 61 | print('Testing {}'.format(op_name)) 62 | 63 | # test empty batch size 64 | _test_primitive( 65 | self, "cpu", 66 | op_name, op_func, 67 | N=0, C_in=16, C_out=32, expand=4, stride=1 68 | ) 69 | 70 | @unittest.skipIf(not TEST_CUDA, "no CUDA detected") 71 | def test_primitives_cuda_empty_batch(self): 72 | ''' Make sures the primitives runs ''' 73 | for op_name, op_func in fbnet_builder.PRIMITIVES.items(): 74 | print('Testing {}'.format(op_name)) 75 | 76 | # test empty batch size 77 | _test_primitive( 78 | self, "cuda", 79 | op_name, op_func, 80 | N=0, C_in=16, C_out=32, expand=4, stride=1 81 | ) 82 | 83 | if __name__ == "__main__": 84 | unittest.main() 85 | -------------------------------------------------------------------------------- /tests/test_feature_extractors.py: -------------------------------------------------------------------------------- 1 | # Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved. 2 | 3 | import unittest 4 | import copy 5 | import torch 6 | # import modules to to register feature extractors 7 | from maskrcnn_benchmark.modeling.backbone import build_backbone # NoQA 8 | from maskrcnn_benchmark.modeling.roi_heads.roi_heads import build_roi_heads # NoQA 9 | from maskrcnn_benchmark.modeling import registry 10 | from maskrcnn_benchmark.structures.bounding_box import BoxList 11 | from maskrcnn_benchmark.config import cfg as g_cfg 12 | from utils import load_config 13 | 14 | # overwrite configs if specified, otherwise default config is used 15 | FEATURE_EXTRACTORS_CFGS = { 16 | } 17 | 18 | # overwrite configs if specified, otherwise default config is used 19 | FEATURE_EXTRACTORS_INPUT_CHANNELS = { 20 | # in_channels was not used, load through config 21 | "ResNet50Conv5ROIFeatureExtractor": 1024, 22 | } 23 | 24 | 25 | def _test_feature_extractors( 26 | self, extractors, overwrite_cfgs, overwrite_in_channels 27 | ): 28 | ''' Make sure roi box feature extractors run ''' 29 | 30 | self.assertGreater(len(extractors), 0) 31 | 32 | in_channels_default = 64 33 | 34 | for name, builder in extractors.items(): 35 | print('Testing {}...'.format(name)) 36 | if name in overwrite_cfgs: 37 | cfg = load_config(overwrite_cfgs[name]) 38 | else: 39 | # Use default config if config file is not specified 40 | cfg = copy.deepcopy(g_cfg) 41 | 42 | in_channels = overwrite_in_channels.get( 43 | name, in_channels_default) 44 | 45 | fe = builder(cfg, in_channels) 46 | self.assertIsNotNone( 47 | getattr(fe, 'out_channels', None), 48 | 'Need to provide out_channels for feature extractor {}'.format(name) 49 | ) 50 | 51 | N, C_in, H, W = 2, in_channels, 24, 32 52 | input = torch.rand([N, C_in, H, W], dtype=torch.float32) 53 | bboxes = [[1, 1, 10, 10], [5, 5, 8, 8], [2, 2, 3, 4]] 54 | img_size = [384, 512] 55 | box_list = BoxList(bboxes, img_size, "xyxy") 56 | out = fe([input], [box_list] * N) 57 | self.assertEqual( 58 | out.shape[:2], 59 | torch.Size([N * len(bboxes), fe.out_channels]) 60 | ) 61 | 62 | 63 | class TestFeatureExtractors(unittest.TestCase): 64 | def test_roi_box_feature_extractors(self): 65 | ''' Make sure roi box feature extractors run ''' 66 | _test_feature_extractors( 67 | self, 68 | registry.ROI_BOX_FEATURE_EXTRACTORS, 69 | FEATURE_EXTRACTORS_CFGS, 70 | FEATURE_EXTRACTORS_INPUT_CHANNELS, 71 | ) 72 | 73 | def test_roi_keypoints_feature_extractors(self): 74 | ''' Make sure roi keypoints feature extractors run ''' 75 | _test_feature_extractors( 76 | self, 77 | registry.ROI_KEYPOINT_FEATURE_EXTRACTORS, 78 | FEATURE_EXTRACTORS_CFGS, 79 | FEATURE_EXTRACTORS_INPUT_CHANNELS, 80 | ) 81 | 82 | def test_roi_mask_feature_extractors(self): 83 | ''' Make sure roi mask feature extractors run ''' 84 | _test_feature_extractors( 85 | self, 86 | registry.ROI_MASK_FEATURE_EXTRACTORS, 87 | FEATURE_EXTRACTORS_CFGS, 88 | FEATURE_EXTRACTORS_INPUT_CHANNELS, 89 | ) 90 | 91 | 92 | if __name__ == "__main__": 93 | unittest.main() 94 | -------------------------------------------------------------------------------- /tests/test_metric_logger.py: -------------------------------------------------------------------------------- 1 | # Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved. 2 | import unittest 3 | 4 | from maskrcnn_benchmark.utils.metric_logger import MetricLogger 5 | 6 | 7 | class TestMetricLogger(unittest.TestCase): 8 | def test_update(self): 9 | meter = MetricLogger() 10 | for i in range(10): 11 | meter.update(metric=float(i)) 12 | 13 | m = meter.meters["metric"] 14 | self.assertEqual(m.count, 10) 15 | self.assertEqual(m.total, 45) 16 | self.assertEqual(m.median, 4) 17 | self.assertEqual(m.avg, 4.5) 18 | 19 | def test_no_attr(self): 20 | meter = MetricLogger() 21 | _ = meter.meters 22 | _ = meter.delimiter 23 | def broken(): 24 | _ = meter.not_existent 25 | self.assertRaises(AttributeError, broken) 26 | 27 | if __name__ == "__main__": 28 | unittest.main() 29 | -------------------------------------------------------------------------------- /tests/test_predictors.py: -------------------------------------------------------------------------------- 1 | # Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved. 2 | 3 | import unittest 4 | import copy 5 | import torch 6 | # import modules to to register predictors 7 | from maskrcnn_benchmark.modeling.backbone import build_backbone # NoQA 8 | from maskrcnn_benchmark.modeling.roi_heads.roi_heads import build_roi_heads # NoQA 9 | from maskrcnn_benchmark.modeling import registry 10 | from maskrcnn_benchmark.config import cfg as g_cfg 11 | from utils import load_config 12 | 13 | 14 | # overwrite configs if specified, otherwise default config is used 15 | PREDICTOR_CFGS = { 16 | } 17 | 18 | # overwrite configs if specified, otherwise default config is used 19 | PREDICTOR_INPUT_CHANNELS = { 20 | } 21 | 22 | 23 | def _test_predictors( 24 | self, predictors, overwrite_cfgs, overwrite_in_channels, 25 | hwsize, 26 | ): 27 | ''' Make sure predictors run ''' 28 | 29 | self.assertGreater(len(predictors), 0) 30 | 31 | in_channels_default = 64 32 | 33 | for name, builder in predictors.items(): 34 | print('Testing {}...'.format(name)) 35 | if name in overwrite_cfgs: 36 | cfg = load_config(overwrite_cfgs[name]) 37 | else: 38 | # Use default config if config file is not specified 39 | cfg = copy.deepcopy(g_cfg) 40 | 41 | in_channels = overwrite_in_channels.get( 42 | name, in_channels_default) 43 | 44 | fe = builder(cfg, in_channels) 45 | 46 | N, C_in, H, W = 2, in_channels, hwsize, hwsize 47 | input = torch.rand([N, C_in, H, W], dtype=torch.float32) 48 | out = fe(input) 49 | yield input, out, cfg 50 | 51 | 52 | class TestPredictors(unittest.TestCase): 53 | def test_roi_box_predictors(self): 54 | ''' Make sure roi box predictors run ''' 55 | for cur_in, cur_out, cur_cfg in _test_predictors( 56 | self, 57 | registry.ROI_BOX_PREDICTOR, 58 | PREDICTOR_CFGS, 59 | PREDICTOR_INPUT_CHANNELS, 60 | hwsize=1, 61 | ): 62 | self.assertEqual(len(cur_out), 2) 63 | scores, bbox_deltas = cur_out[0], cur_out[1] 64 | self.assertEqual( 65 | scores.shape[1], cur_cfg.MODEL.ROI_BOX_HEAD.NUM_CLASSES) 66 | self.assertEqual(scores.shape[0], cur_in.shape[0]) 67 | self.assertEqual(scores.shape[0], bbox_deltas.shape[0]) 68 | self.assertEqual(scores.shape[1] * 4, bbox_deltas.shape[1]) 69 | 70 | def test_roi_keypoints_predictors(self): 71 | ''' Make sure roi keypoint predictors run ''' 72 | for cur_in, cur_out, cur_cfg in _test_predictors( 73 | self, 74 | registry.ROI_KEYPOINT_PREDICTOR, 75 | PREDICTOR_CFGS, 76 | PREDICTOR_INPUT_CHANNELS, 77 | hwsize=14, 78 | ): 79 | self.assertEqual(cur_out.shape[0], cur_in.shape[0]) 80 | self.assertEqual( 81 | cur_out.shape[1], cur_cfg.MODEL.ROI_KEYPOINT_HEAD.NUM_CLASSES) 82 | 83 | def test_roi_mask_predictors(self): 84 | ''' Make sure roi mask predictors run ''' 85 | for cur_in, cur_out, cur_cfg in _test_predictors( 86 | self, 87 | registry.ROI_MASK_PREDICTOR, 88 | PREDICTOR_CFGS, 89 | PREDICTOR_INPUT_CHANNELS, 90 | hwsize=14, 91 | ): 92 | self.assertEqual(cur_out.shape[0], cur_in.shape[0]) 93 | self.assertEqual( 94 | cur_out.shape[1], cur_cfg.MODEL.ROI_BOX_HEAD.NUM_CLASSES) 95 | 96 | 97 | if __name__ == "__main__": 98 | unittest.main() 99 | -------------------------------------------------------------------------------- /tests/test_rpn_heads.py: -------------------------------------------------------------------------------- 1 | # Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved. 2 | 3 | import unittest 4 | import copy 5 | import torch 6 | # import modules to to register rpn heads 7 | from maskrcnn_benchmark.modeling.backbone import build_backbone # NoQA 8 | from maskrcnn_benchmark.modeling.rpn.rpn import build_rpn # NoQA 9 | from maskrcnn_benchmark.modeling import registry 10 | from maskrcnn_benchmark.config import cfg as g_cfg 11 | from utils import load_config 12 | 13 | 14 | # overwrite configs if specified, otherwise default config is used 15 | RPN_CFGS = { 16 | } 17 | 18 | 19 | class TestRPNHeads(unittest.TestCase): 20 | def test_build_rpn_heads(self): 21 | ''' Make sure rpn heads run ''' 22 | 23 | self.assertGreater(len(registry.RPN_HEADS), 0) 24 | 25 | in_channels = 64 26 | num_anchors = 10 27 | 28 | for name, builder in registry.RPN_HEADS.items(): 29 | print('Testing {}...'.format(name)) 30 | if name in RPN_CFGS: 31 | cfg = load_config(RPN_CFGS[name]) 32 | else: 33 | # Use default config if config file is not specified 34 | cfg = copy.deepcopy(g_cfg) 35 | 36 | rpn = builder(cfg, in_channels, num_anchors) 37 | 38 | N, C_in, H, W = 2, in_channels, 24, 32 39 | input = torch.rand([N, C_in, H, W], dtype=torch.float32) 40 | LAYERS = 3 41 | out = rpn([input] * LAYERS) 42 | self.assertEqual(len(out), 2) 43 | logits, bbox_reg = out 44 | for idx in range(LAYERS): 45 | self.assertEqual( 46 | logits[idx].shape, 47 | torch.Size([ 48 | input.shape[0], num_anchors, 49 | input.shape[2], input.shape[3], 50 | ]) 51 | ) 52 | self.assertEqual( 53 | bbox_reg[idx].shape, 54 | torch.Size([ 55 | logits[idx].shape[0], num_anchors * 4, 56 | logits[idx].shape[2], logits[idx].shape[3], 57 | ]), 58 | ) 59 | 60 | 61 | if __name__ == "__main__": 62 | unittest.main() 63 | -------------------------------------------------------------------------------- /tests/utils.py: -------------------------------------------------------------------------------- 1 | from __future__ import absolute_import, division, print_function, unicode_literals 2 | 3 | # Set up custom environment before nearly anything else is imported 4 | # NOTE: this should be the first import (no not reorder) 5 | from maskrcnn_benchmark.utils.env import setup_environment # noqa F401 isort:skip 6 | import env_tests.env as env_tests 7 | 8 | import os 9 | import copy 10 | 11 | from maskrcnn_benchmark.config import cfg as g_cfg 12 | 13 | 14 | def get_config_root_path(): 15 | return env_tests.get_config_root_path() 16 | 17 | 18 | def load_config(rel_path): 19 | ''' Load config from file path specified as path relative to config_root ''' 20 | cfg_path = os.path.join(env_tests.get_config_root_path(), rel_path) 21 | return load_config_from_file(cfg_path) 22 | 23 | 24 | def load_config_from_file(file_path): 25 | ''' Load config from file path specified as absolute path ''' 26 | ret = copy.deepcopy(g_cfg) 27 | ret.merge_from_file(file_path) 28 | return ret 29 | -------------------------------------------------------------------------------- /tools/cityscapes/instances2dict_with_polygons.py: -------------------------------------------------------------------------------- 1 | #!/usr/bin/python 2 | # 3 | # Convert instances from png files to a dictionary 4 | # This files is created according to https://github.com/facebookresearch/Detectron/issues/111 5 | 6 | from __future__ import print_function, absolute_import, division 7 | import os, sys 8 | 9 | sys.path.append( os.path.normpath( os.path.join( os.path.dirname( __file__ ) , '..' , 'helpers' ) ) ) 10 | from csHelpers import * 11 | 12 | # Cityscapes imports 13 | from cityscapesscripts.evaluation.instance import * 14 | from cityscapesscripts.helpers.csHelpers import * 15 | import cv2 16 | from maskrcnn_benchmark.utils import cv2_util 17 | 18 | 19 | def instances2dict_with_polygons(imageFileList, verbose=False): 20 | imgCount = 0 21 | instanceDict = {} 22 | 23 | if not isinstance(imageFileList, list): 24 | imageFileList = [imageFileList] 25 | 26 | if verbose: 27 | print("Processing {} images...".format(len(imageFileList))) 28 | 29 | for imageFileName in imageFileList: 30 | # Load image 31 | img = Image.open(imageFileName) 32 | 33 | # Image as numpy array 34 | imgNp = np.array(img) 35 | 36 | # Initialize label categories 37 | instances = {} 38 | for label in labels: 39 | instances[label.name] = [] 40 | 41 | # Loop through all instance ids in instance image 42 | for instanceId in np.unique(imgNp): 43 | if instanceId < 1000: 44 | continue 45 | instanceObj = Instance(imgNp, instanceId) 46 | instanceObj_dict = instanceObj.toDict() 47 | 48 | #instances[id2label[instanceObj.labelID].name].append(instanceObj.toDict()) 49 | if id2label[instanceObj.labelID].hasInstances: 50 | mask = (imgNp == instanceId).astype(np.uint8) 51 | contour, hier = cv2_util.findContours( 52 | mask.copy(), cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_NONE) 53 | 54 | polygons = [c.reshape(-1).tolist() for c in contour] 55 | instanceObj_dict['contours'] = polygons 56 | 57 | instances[id2label[instanceObj.labelID].name].append(instanceObj_dict) 58 | 59 | imgKey = os.path.abspath(imageFileName) 60 | instanceDict[imgKey] = instances 61 | imgCount += 1 62 | 63 | if verbose: 64 | print("\rImages Processed: {}".format(imgCount), end=' ') 65 | sys.stdout.flush() 66 | 67 | if verbose: 68 | print("") 69 | 70 | return instanceDict 71 | 72 | def main(argv): 73 | fileList = [] 74 | if (len(argv) > 2): 75 | for arg in argv: 76 | if ("png" in arg): 77 | fileList.append(arg) 78 | instances2dict_with_polygons(fileList, True) 79 | 80 | if __name__ == "__main__": 81 | main(sys.argv[1:]) 82 | -------------------------------------------------------------------------------- /tools/test_net.py: -------------------------------------------------------------------------------- 1 | # Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved. 2 | # Set up custom environment before nearly anything else is imported 3 | # NOTE: this should be the first import (no not reorder) 4 | from maskrcnn_benchmark.utils.env import setup_environment # noqa F401 isort:skip 5 | 6 | import argparse 7 | import os 8 | 9 | import torch 10 | from maskrcnn_benchmark.config import cfg 11 | from maskrcnn_benchmark.data import make_data_loader 12 | from maskrcnn_benchmark.engine.inference import inference 13 | from maskrcnn_benchmark.modeling.detector import build_detection_model 14 | from maskrcnn_benchmark.utils.checkpoint import DetectronCheckpointer 15 | from maskrcnn_benchmark.utils.collect_env import collect_env_info 16 | from maskrcnn_benchmark.utils.comm import synchronize, get_rank 17 | from maskrcnn_benchmark.utils.logger import setup_logger 18 | from maskrcnn_benchmark.utils.miscellaneous import mkdir 19 | 20 | 21 | def main(): 22 | parser = argparse.ArgumentParser(description="PyTorch Object Detection Inference") 23 | parser.add_argument( 24 | "--config-file", 25 | default="/private/home/fmassa/github/detectron.pytorch_v2/configs/e2e_faster_rcnn_R_50_C4_1x_caffe2.yaml", 26 | metavar="FILE", 27 | help="path to config file", 28 | ) 29 | parser.add_argument("--local_rank", type=int, default=0) 30 | parser.add_argument( 31 | "opts", 32 | help="Modify config options using the command-line", 33 | default=None, 34 | nargs=argparse.REMAINDER, 35 | ) 36 | 37 | args = parser.parse_args() 38 | 39 | num_gpus = int(os.environ["WORLD_SIZE"]) if "WORLD_SIZE" in os.environ else 1 40 | distributed = num_gpus > 1 41 | 42 | if distributed: 43 | torch.cuda.set_device(args.local_rank) 44 | torch.distributed.init_process_group( 45 | backend="nccl", init_method="env://" 46 | ) 47 | synchronize() 48 | 49 | cfg.merge_from_file(args.config_file) 50 | cfg.merge_from_list(args.opts) 51 | cfg.freeze() 52 | 53 | save_dir = "" 54 | logger = setup_logger("maskrcnn_benchmark", save_dir, get_rank()) 55 | logger.info("Using {} GPUs".format(num_gpus)) 56 | logger.info(cfg) 57 | 58 | logger.info("Collecting env info (might take some time)") 59 | logger.info("\n" + collect_env_info()) 60 | 61 | model = build_detection_model(cfg) 62 | model.to(cfg.MODEL.DEVICE) 63 | 64 | output_dir = cfg.OUTPUT_DIR 65 | checkpointer = DetectronCheckpointer(cfg, model, save_dir=output_dir) 66 | _ = checkpointer.load(cfg.MODEL.WEIGHT) 67 | 68 | iou_types = ("bbox",) 69 | if cfg.MODEL.MASK_ON: 70 | iou_types = iou_types + ("segm",) 71 | if cfg.MODEL.KEYPOINT_ON: 72 | iou_types = iou_types + ("keypoints",) 73 | output_folders = [None] * len(cfg.DATASETS.TEST) 74 | dataset_names = cfg.DATASETS.TEST 75 | if cfg.OUTPUT_DIR: 76 | for idx, dataset_name in enumerate(dataset_names): 77 | output_folder = os.path.join(cfg.OUTPUT_DIR, "inference", dataset_name) 78 | mkdir(output_folder) 79 | output_folders[idx] = output_folder 80 | data_loaders_val = make_data_loader(cfg, is_train=False, is_distributed=distributed) 81 | for output_folder, dataset_name, data_loader_val in zip(output_folders, dataset_names, data_loaders_val): 82 | inference( 83 | model, 84 | data_loader_val, 85 | dataset_name=dataset_name, 86 | iou_types=iou_types, 87 | box_only=False if cfg.MODEL.RETINANET_ON else cfg.MODEL.RPN_ONLY, 88 | device=cfg.MODEL.DEVICE, 89 | expected_results=cfg.TEST.EXPECTED_RESULTS, 90 | expected_results_sigma_tol=cfg.TEST.EXPECTED_RESULTS_SIGMA_TOL, 91 | output_folder=output_folder, 92 | ) 93 | synchronize() 94 | 95 | 96 | if __name__ == "__main__": 97 | main() 98 | --------------------------------------------------------------------------------