First, great work here! I've been digging through the code, and it's been incredibly helpful learning how these coefficient domain convolutions work. On a slightly unrelated note, in the ReLU implementation here, why is it possible to avoid rescaling? Thanks!
Hi,
First, great work here! I've been digging through the code, and it's been incredibly helpful learning how these coefficient domain convolutions work. On a slightly unrelated note, in the ReLU implementation here, why is it possible to avoid rescaling? Thanks!