Hello all,
I want to implement my custom convolution Function for custom gradient.
So, I need to calculate original gradient of convolution.
How can I implement below code?
class MyConvolution(Function):
@staticmethod
def forward(ctx, x, w):
ctx.save_for_backward(x, w)
out = nn.functional.conv2d(x, w)
return out
@staticmethod
def backward(ctx, grad_output):
x, w, = ctx.saved_tensors
return **???, ???**