Open daiisuke27 opened 4 years ago
hello,
1/ which training set do you use ? I never get nan
2/ my loss fucntion returns only xy_loss + wh_loss +obj_loss +class_loss , so you can't get only obj_loss or class_loss, why do you need them ?
in fact, regrading the nan issue, i have the same error when i activate GPU. but in in this case i have also error to allocate memory for convolution redzone , maybe it is the root cause of the nan
I have comitted a fix, you can try again
Hello, i am running your code on Colab.
Thank you.