Both tf1.14 and tf2.0 version hang during training
yuanli12139 opened this issue · 3 comments
yuanli12139 commented
Hi @610265158, I tried to train FaceBoxes detectors using your implementation. However, both tf1.14 and tf2.0 version hang after training started several hundred iterations. No error messages but losses stopped updating. Do you have any idea what might be the cause? Thank you so much!
610265158 commented
It should be the data provider problem, some err may happened there , i guess.
Which dataset do you use?
yuanli12139 commented
I am using Wider Face despite that I am training 1-channel grayscale models.
610265158 commented
Do you have problem with raw 3-channel data?
You could catch some exception in the data provider， and to find out the problem.