HELP I'm trying the WGAN notebook, but it only produces noise that's meaningless

I’m trying fastai’s WGAN notebook.

The part I changed in the original code was just to get my data instead of bedroom data

path = untar_data(URLs.LSUN_BEDROOMS)
path = Path(f'/home/mskang/sungho/src/data/source/zombies')

But no matter how much I train WGAN, it produces only noise!

When I used the bedroom data in the same environment, it worked fine.

I don’t know what makes the difference. Even training with only one data produces only noise (in my prediction, I thought it would mimic that single data).

Perhaps the problem is the difference in the number of data? Couldn’t it be overwhelmed by the amount of training?


Please let me know if anyone knows a hint that I can try!

My data is 123 images.