This repo follow by https://github.com/bubbliiiing/faster-rcnn-pytorch, I make it easy to study.
frcnn_all.py = stage1(rpn) +stage2(classifier) You should deal with the coordinates of the propsals box(rois) and its corrections from stage2. There's a problem here that I need to deal with is that how to understand DecodeBox(). But it didn't influence you predict the model.
https://drive.google.com/file/d/1bXLxg0jkZDZeYoOQARE4C1M4nnQGXtU3/view?usp=sharing
- Test the model on VOC datasets
- Add the train code.