版本使用的是 torch1.4.0
cuda 使用的是cuda10.0
export CUDA_HOME=/usr/local/cuda
fist:
install dependency:
conda create --name rotated_maskrcnn python=3.6
conda activate rotated_maskrcnn
conda install ipython
pip install ninja yacs cython matplotlib tqdm opencv-python
conda install pytorch1.4.0 torchvision0.5.0 cudatoolkit=10.0 -c pytorch
git clone https://github.com/cocodataset/cocoapi.git
cd cocoapi/PythonAPI
python setup.py build_ext install
git reset --hard 3fe10b5597ba14a748ebb271a6ab97c09c5701ac
git clone https://github.com/NVIDIA/apex.git
cd apex
python setup.py install --cuda_ext --cpp_ext
git clone https://github.com/mrlooi/rotated_maskrcnn.git
cd rotated_maskrcnn
python setup.py build develop
http://datasets.d2.mpi-inf.mpg.de/hosang17cvpr/coco_minival2014.tar.gz
mkdir -p datasets/coco
ln -s /path_to_coco_dataset/annotations datasets/coco/annotations
ln -s /path_to_coco_dataset/train2014 datasets/coco/train2014
ln -s /path_to_coco_dataset/test2014 datasets/coco/test2014
ln -s /path_to_coco_dataset/val2014 datasets/coco/val2014
change the torch.uint8 to torch.bool:
maskrcnn_benchmark/structures/segmentation_mask.py
maskrcnn_benchmark/modeling/balanced_positive_negative_sampler.py
python tools/train_net.py --config-file “configs/rotated/e2e_ms_rcnn_R_50_FPN_1x.yaml” SOLVER.IMS_PER_BATCH 2 SOLVER.BASE_LR 0.0025 SOLVER.MAX_ITER 720000 SOLVER.STEPS “(480000, 640000)” TEST.IMS_PER_BATCH 1
train your own dataset:
labelme https://github.com/wkentaro/labelme/tree/main/examples/instance_segmentation
./labelme2coco.py data_annotated data_dataset_coco --labels labels.txt
Convert your own dataset to COCO annotation format (a json file). i use the labelme to do it.
Add the path of the dataset image directory and coco-format annotation file (from Step 1) into maskrcnn_benchmark/config/paths_catalog.py remember change the num_class
In your .yaml config file (e.g. “configs/rotated/e2e_ms_rcnn_R_50_FPN_1x.yaml”), change the DATASETS.TRAIN value to the stuff you added in paths_catalog.py.
change the MODEL.ROI_HEADS.NMS to 0.3
