This repo is implemented based on detectron2 and CenterNet-better
- Python >= 3.6
- PyTorch >= 1.4
- torchvision that matches the PyTorch installation.
- OpenCV
- pycocotools
pip install cython; pip install 'git+https://github.com/cocodataset/cocoapi.git#subdirectory=PythonAPI'
- GCC >= 4.9
gcc --version
- detectron2
pip install -U 'git+https://github.com/facebookresearch/detectron2.git'
python train_net.py --num-gpus 8 --config-file configs/centernet_r_18_C4_1x.yaml
python train_net.py --num-gpus 8 --config-file configs/centernet_r_18_C4_1x.yaml --eval-only MODEL.WEIGHTS model_0007999.pth
This repo use less training time to get a better performance, it nearly spend half training time and get 1~2 pts higher mAP compared with the old repo. Here is the table of performance.
Backbone ResNet-50
Code | mAP |
---|---|
ours | |
centernet-better | 35.1 |
Backbone ResNet-18
Code | mAP |
---|---|
ours | 29.7 |
centernet-better | 29.8 |
- Support DLA backbone
- Support Hourglass backbone
- Support KeyPoints dataset