I'm training on my own dataset. I found ransac_voting_layer_v3 generate wrong 2d keypoints even using the ground truth vertex in the __getitem__ function of lib/datasets/custom/pvnet.py. However, using the ransac_voting_layer can generate correct results. Could you help explain why this happen?
I'm training on my own dataset. I found
ransac_voting_layer_v3
generate wrong 2d keypoints even using the ground truthvertex
in the__getitem__
function oflib/datasets/custom/pvnet.py
. However, using theransac_voting_layer
can generate correct results. Could you help explain why this happen?