Does I need to rewrite the backward function which a part of the forward net like this?
pool = []
for i in range(n):
pool.append(F.adaptive_avg_pool2d(img, 7))
pool = torch.cat(pool, 0)
Does I need to rewrite the backward function which a part of the forward net like this?
pool = []
for i in range(n):
pool.append(F.adaptive_avg_pool2d(img, 7))
pool = torch.cat(pool, 0)
normally you don’t need to write backward at all unless you are writing a custom autograd function.