Convolution operation without the final summation

4

In that case, the groups argument should yield the expected results using groups=in_channels:

conv = nn.Conv2d(3, 3, 3, groups=3)
x = torch.randn(10, 3, 64, 64)
output = conv(x)
1 Like