GengDavid/pytorch-cpn

some question about the human detector.

zhwzhao opened this issue · 0 comments

in this paper, I see that CPN used the image cropped in raw image with the bounding box of FPN as input. Where is the FPN in your reimplement ?