How to set the size of weights memory when convolution‘s groups is not equal to 1.

Question

How to set the size of weights memory when convolution‘s groups is not equal to 1.

Closed this issue 6 years ago · 1 comments

Hi,

As described in the example MNIST network, the size of convolution's weights memory is set to { out_channels, in_channels, kernel_size, kernel_size} when groups is 1. My questions is when groups is not 1, the size is still the same?

Thanks.

Answer 1 · 2019-05-21T07:02:14.000Z

Already found the answer in test file.