rardz

From the code, it seems the optimized 'noise constant' in the first step will be abandoned, and in the next pti step, the model just use another irrelevant constant noise....

Seems like a hastily modified version from some of stylegan's modules.

I also have the same doubts. What is the reason for the disorder of the numbering with 50001 ? Is there any way to fix it? Hope the author can...

I think your code `bucketsampler` should be an iterator that randomly samples from all the tarfiles of every bucket. In your demo code, the data is merely fetched randomly, shuffled...