SPI-GAN: Denoising Diffusion GANs with Straight-Path Interpolations

Jinsung Jeon, Noseong Park

Abstract: Score-based generative models (SGMs) are a recently proposed paradigm for deep generative tasks and now show the state-of-the-art sampling performance. It is known that the original SGM design solves the two problems of the generative trilemma: i) sampling quality, and ii) sampling diversity. However, the last problem of the trilemma was not solved, i.e., their training/sampling complexity is notoriously high. To this end, combining SGMs with simpler models, e.g., generative adversarial networks (GANs), is gathering much attention currently. We present an enhanced denoising method using GANs, called straight-path interpolation GAN (SPI-GAN), which drastically reduces the sampling time while achieving as high sampling quality and diversity as SGMs. Our SPI-GAN can be compared to the state-of-the-art shortcut-based denoising method using GANs, called denoising diffusion GAN (DD-GAN). However, our method corresponds to an extreme method that does not use any intermediate shortcut information of the reverse SDE path, in which case DD-GAN ($K=1$) fails to obtain good results. Nevertheless, our straight-path interpolation method greatly stabilizes the overall training process. As a result, SPI-GAN is one of the best-balanced models in terms of the sampling quality/diversity/time for CIFAR-10, CelebA-HQ-256, and LSUN-Church-256.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics

Submission Guidelines: Yes

Please Choose The Closest Area That Your Submission Falls Into: Generative models