Can Gated Convolutions be used in a GAN?

[Mayur28]

Hi,

I have read the gated convolution paper (and reviewed the results achieved by the implementation in this repo) and the results are fascinating!
I just have a few questions that I was hoping to clarify please:

1. How necessary was it to include batch normalization in the implementation in the repo (in terms of the difference in quality of the results achieved compared to the original implementation)?
2. I've only come across papers and implementations where gated convolutions are implemented using the contextual attention (coarse and refinement network) as a baseline/foundation. I just wanted to find out whether gated convolutions can be employed in other (more general) networks, such as substituting vanilla conv layers with gated convolutions?

Thanks