search

longcw / yolo2-pytorch

YOLOv2 in PyTorch
1.54k stars 421 forks source link

_ious[cell_ind, a, :] = ious_reshaped[cell_ind, a, i] #64

Open 22wei22 opened 6 years ago

22wei22 commented 6 years ago

Ask for help,in your code,you use _ious[cell_ind, a, :] = ious_reshaped[cell_ind, a, i], I think _ious[cell_ind, a, :] = anchor_ious[a, i]

Erotemic commented 6 years ago

No, the predicted IOU should be compared to the IOU with the groundtruth box, not the anchor box. I'm vetting this code fairly extensively right now, and I believe this is right.

See: https://github.com/longcw/yolo2-pytorch/issues/19