search

FoundationVision / LlamaGen

Autoregressive Model Beats Diffusion: 🦙 Llama for Scalable Image Generation
https://arxiv.org/abs/2406.06525
MIT License
888 stars 32 forks source link

Questions about the discriminator #35

Open Doctor-James opened 1 week ago

Doctor-James commented 1 week ago

Follow issue https://github.com/FoundationVision/LlamaGen/issues/27 I would like to ask if the discriminator's logit_real and logit_fake are always similar, and when they are both positive, the discriminator_adv_loss will be negative. Is such a discriminator beneficial to the generator?

PeizeSun commented 3 days ago

Hi~ I understand this is magic. The adversarial process can’t be dominated by either generator or discriminator. Also, I highly recommend to run other GAN codebases to see this magic phenomena.