Thanks for the nice tutorial. I was able to train my custom datasets but got loss=Nan.
I also decreased the learning rate but couldn't fix the problem?

Any suggestions?