search

pythonlessons / TensorFlow-2.x-YOLOv3

YOLOv3 implementation in TensorFlow 2.3.1
https://pylessons.com/
MIT License
605 stars 327 forks source link

ciou_loss #144

Closed lihualilee closed 3 years ago

lihualilee commented 3 years ago

when i use ciou_loss，the loss number always is nan? image about ar_gt = boxes2[..., 2] / boxes2[..., 3]，We're using division, but we're going to have a lot of zeros in the denominator

pythonlessons commented 3 years ago

Yes, ciou is not working

WillLiGitHub commented 2 years ago

change the function bbox_ciou(): ar_gt = boxes2[..., 2] / (boxes2[..., 3] + 0.000001) ar_pred = boxes1[..., 2] / (boxes1[..., 3] + 0.000001)