Performance on MNIST - Githubissues

lucidrains / stylegan2-pytorch

Simplest working implementation of Stylegan2, state of the art generative adversarial network, in Pytorch. Enabling everyone to experience disentanglement

https://thispersondoesnotexist.com

MIT License

3.72k stars 588 forks source link

Performance on MNIST #267

Open koriavinash1 opened 2 years ago

koriavinash1 commented 2 years ago

@lucidrains Thanks for this repository.

I'm trying this script for generating Morpho-MNIST digits (MNIST with tweaking morphology like thickness and intensity), image size 32x32, 3 channels.

Even after training for approximately upto 200k I'm unable to get FID < 60. I'm attaching some of the generated images, it would be really helpful if you could provide some insights on this.

Thank you, Avinash 196500-ema 195000-ema

exlix96 commented 3 months ago

hi, As you are using MNIST dataset and the inputs are in grayscale, I wonder why your image is getting 3 channel as the input?