- An implementation of Faster R-CNN in PyTorch.
- Only python source. These is no need to compile nms and roialign cpp
- Comment in many functions
- You can debug any line in cpu mode
- cd faster_rcnn_pytorch
- ./install_data.sh
- python3 ./train.py --cuda True
- python3 ./train.py --cuda True --resume True
- python3 ./infer.py --cuda True
- GeForce GTX 1650 4GB
- CUDA version 10.2
- Resnet-101
- Train 2.3 frames per second
- Infer 5 frames per second
- https://drive.google.com/file/d/1dPoxbqVq3RswQx7T4nn19_qEHnMK_0Ep
- Copy to ./checkpoint
You can compare with simple-faster-rcnn that can debug any line in cpu mode