When I was debugging, I noticed the grad_fn for a tensor is ‘UpsampleBilinear2DBackward’. I was confused that can nn.Upsample do differentiation? I know that nn.Conv2d and nn.ConvTranspose2d() can do differentiation and get the gradients because they have parameters.
Besides, how can I take a look that how ops computes gradients in my IDE (PyCharm) or at somewhere else?