Note that, for mmcv, please install the version compatible to your cuda version. You can find the install guide here
conda create -n dropout_rashomon_det python==3.8.11
conda activate dropout_rashomon_det
pip install -r requirements.txt
pip install mmcv==2.0.1 -f https://download.openmmlab.com/mmcv/dist/cu117/torch2.0/index.html
git clone https://github.com/open-mmlab/mmdetection.git
cd mmdetection
git checkout f78af7785ada87f1ced75a2313746e4ba3149760
pip install -e .
cp -f ../test.py tools/test.py
Please follow the instructions from mmdetection to setup MS COCO dataset and get pretrained models (YoloV3 and MaskRCNN) from their GitHub and put in checkpoints folder.
From mmdetection directory, download checkpoints.
mkdir -p checkpoints
cd checkpoints
wget https://download.openmmlab.com/mmdetection/v2.0/mask_rcnn/mask_rcnn_r50_fpn_2x_coco/mask_rcnn_r50_fpn_2x_coco_bbox_mAP-0.392__segm_mAP-0.354_20200505_003907-3e542a40.pth
wget https://download.openmmlab.com/mmdetection/v2.0/yolo/yolov3_d53_320_273e_coco/yolov3_d53_320_273e_coco-421362b6.pth
Go back to detecton folder, use the generate_run_all.py to generate bash script to control type of dropout and dropout rate.
Then, copy the generated bash scripts and analysis python file (predicitive_multiplicitiy_analysis.py) into mmdetection folder to run detection results with different settings.
YoloV3 results are stored in rashomon folder while MaskRCNN results are located in rashomon_maskrcnn folder.
python3 generate_run_all.py
cp run_all_yolov3.sh mmdetection/
cp run_all_maskrcnn.sh mmdetection/
cp predicitive_multiplicitiy_analysis.py mmdetection/
cd mmdetection/
bash run_all_yolov3.sh
bash run_all_maskrcnn.sh
python3 predicitive_multiplicitiy_analysis.py yolov3
python3 predicitive_multiplicitiy_analysis.py maskrcnn