StyleGAN not training

Course Q&A Generative Adversarial Networks (GANS) Apply Generative Adversarial Networks
1

Hi

After finishing the GAN specialization, I set out to train some of the models on easier datasets, I first trained the proGAN on the CIFAR-10 dataset and it achieved considerable success. Then I followed onto training the styleGAN on the same dataset and the results were not satisfactory at all. Despite using W-GAN-GP loss the styleGAN fell into model collapse (of blurry garbage) while training even for the 4x4 resolution (among other resolutions). I’m pretty sure my implementation is correct and so is the progressive growing, then why is this model not training at all? If needed I’ll share my code, I’m not doing so already since it’s a lot of lines.

Can someone assist with me this please?

2

Hi @Mohammed_Ali4!
Firstly, sorry for the late reply.
Are you using the official StyleGAN repository? GitHub - NVlabs/stylegan: StyleGAN - Official TensorFlow Implementation this is the link to TF implementation but there’s also Pytorch repository. In case you’re using the official repository and still experiencing issues please try raising an issue in that repository.
Thank you and hope to hear back from you.