Question about self._predictions["bbox_pred"]

endernewton / tf-faster-rcnn

Tensorflow Faster RCNN for Object Detection

https://arxiv.org/pdf/1702.02138.pdf

MIT License

3.65k stars 1.57k forks source link

Question about self._predictions["bbox_pred"] #321

Closed mklissa closed 6 years ago

mklissa commented 6 years ago

Hi there, thanks for your code, its very useful! I was trying to understand how bbox_pred affects the RoI, if it affects it at all? I can see that rpn_bbox_pred affects the RoIs through self._proposal_layer(rpn_cls_prob, rpn_bbox_pred, "rois"). However, I don't see something similar for bbox_pred. Why are we calculating the loss for this variable if doesn't affect the RoI during training?