Thx 4 ur reply. the behavior was that my model wasn’t learning anything, the loss wasn’t changing a lot and the accuracy(measured per epoch) wasn’t even changing. So that’s y I thought the gradients wasn’t computed correctly.
However, I solved it already, it’s due to a problem in other part of my model; in fact, it was because I used too many Stochastic pooling and that the model was learning very slowly.
Thank you so much:smiley: