futurexiang / soda Goto Github PK

Hi FutureXiang,
In the SODA paper, the convolutional UNet have 2m + 1 activation layers h_i. The author partition the latent vector z into m + 1 sections, and use each z_i to modulate the respective pair or layers (h_i, h_{2m−i}). Do you seem to have used the entire latent vector z in each decoder layer in your implementation, or did I make a mistake?

Hey there, thank you very much for working on this repo! I highly appreciate it!

I wanted to ask if you could please add in the title of the github page (the readme file) that his is unofficial implementation please?
I've seen it's mentioned in the about section but that might be quite hard to notice.

I'm also planning to release an official version soon.

Thanks a lot!

• Drew