CANet-MMDetection

CANet based on MMDetection

• Paper: CANet: Contextual information and spatial attention based network for detecting small defects in manufacturing industry
• Author: Xiuquan Hou, Meiqin Liu, Senlin Zhang, Ping Wei, Badong Chen.

💖 If our CANet is helpful to your researches or projects, please help star this repository. Thanks! 🤗

Code is tested with python=3.10, pytorch=1.12.1, torchvision=0.13.1, mmcv-full=1.7.0, mmdet=2.25.2, other version may also work.

• Change Directory to CANet-MMDetection:

  cd CANet-MMDetection

• Install Pytorch:

  conda install pytorch==1.12.1 torchvision==0.13.1 torchaudio==0.12.1 cudatoolkit=11.3 -c pytorch
  

• Install MMCV and MMDetection:

  pip install -U openmim
  mim install mmcv-full==1.7.0
  pip install -v -e .

• Prepare datasets and put them in the data/ directory:

    data/
     ├─coco/
     │  ├── train2017/
     │  ├── val2017/
     │  └── annotations/
     │
     ├─neu_det_coco/
     │  ├── train2017/
     │  ├── val2017/
     │  └── annotations/
     │
     └─VOCdevkit
        ├── VOC2007/
        └── VOC2012/

• Train:

  # train CANet on VOC dataset
  python tools/train.py configs/canet/canet_r50_fpn_1x_voc0712.py
  # train CANet on COCO dataset
  python tools/train.py configs/canet/canet_r50_fpn_1x_coco.py
  # train CANet on NEU-DET dataset
  python tools/train.py configs/canet/canet_r50_fpn_1x_neu_det_coco.py

• Test:

  # test CANet on VOC dataset
  python tools/test.py configs/canet/canet_r50_fpn_1x_voc0712.py <path/to/checkpoint.pth> --eval mAP
  # test CANet on COCO dataset
  python tools/test.py configs/canet/canet_r50_fpn_1x_coco.py <path/to/checkpoint.pth> --eval bbox
  # test CANet on NEU-DET dataset
  python tools/test.py configs/canet/canet_r50_fpn_1x_neu_det_coco.py <path/to/checkpoint.pth> --eval bbox

• See documents under the docs/ directory for other usage.