No speedup. Training is slower using Lego filters #2
Description
The current given implementation does not provide any speed up in training. The usage of lego filters actually slows down the training process compared to regular convolutional filters. Can you explain why this is happening?