Install

1. conda install pytorch=1.6.0 cudatoolkit=10.1 torchvision -c pytorch

2. pip install mmcv-full -f https://download.openmmlab.com/mmcv/dist/cu101/torch1.6.0/index.html

3. git clone https://github.com/Pstage-Segmentation-Detection/mmdetection_trash.git

4. cd mmdetection_trash

5. pip install -r requirements.txt

6. pip install -v -e .

7. apt-get install libgl1-mesa-glx

8. jupyter nbextension enable --py widgetsnbextension

notebook

train

faster_rcnn_train.ipynb

inference

faster_rcnn_inference.ipynb

python

train

python tools/train.py configs/trash/faster_rcnn/faster_rcnn_r50_fpn_1x_trash.py

inference

python tools/test.py configs/trash/faster_rcnn/faster_rcnn_r50_fpn_1x_trash.py work_dirs/faster_rcnn_r50_fpn_1x_trash/epoch_12.pth --out work_dirs/faster_rcnn_r50_fpn_1x_trash/epoch_12.pkl

make submission

python pkl_to_submission.py --pkl work_dirs/faster_rcnn_r50_fpn_1x_trash/epoch_12.pkl --csv submission.csv

submission file

mmdetection_trash > submission.csv