keras-yolo3 loss is descending,but the val loss=nan???

loss is descending,but the val loss=nan???

Open hexin6688 opened this issue 4 years ago • 4 comments

loss is descending,but the val loss=nan,anybody have face this problem?

Apr 20 '20 12:04 hexin6688

Yes I faced it too..when I tried with tiny yolo model and custom anchors

Apr 22 '20 11:04 kartikwar

I did face this issue for val_loss when disabled the random flag on the validation data generator

Apr 29 '20 08:04 BBarbosa

when trainning on the first stage,the val loss=nan,but when training on the second stage(unfreezing) the val loss starts to descending.

Apr 30 '20 09:04 hexin6688

Have you solved your problem? What is the reason?

Mar 10 '21 01:03 Zhang-KL