Improved Vector Quantized Diffusion Models
This work addresses quality issues in text-to-image synthesis for AI applications, representing an incremental improvement over existing VQ-Diffusion methods.
The paper tackles the problem of low-quality and weakly correlated text-to-image generation in Vector Quantized Diffusion (VQ-Diffusion) models by proposing improved sampling strategies, resulting in significant gains such as reducing the FID score on MSCOCO from 13.86 to 8.44 and on ImageNet from 11.89 to 4.83.
Vector quantized diffusion (VQ-Diffusion) is a powerful generative model for text-to-image synthesis, but sometimes can still generate low-quality samples or weakly correlated images with text input. We find these issues are mainly due to the flawed sampling strategy. In this paper, we propose two important techniques to further improve the sample quality of VQ-Diffusion. 1) We explore classifier-free guidance sampling for discrete denoising diffusion model and propose a more general and effective implementation of classifier-free guidance. 2) We present a high-quality inference strategy to alleviate the joint distribution issue in VQ-Diffusion. Finally, we conduct experiments on various datasets to validate their effectiveness and show that the improved VQ-Diffusion suppresses the vanilla version by large margins. We achieve an 8.44 FID score on MSCOCO, surpassing VQ-Diffusion by 5.42 FID score. When trained on ImageNet, we dramatically improve the FID score from 11.89 to 4.83, demonstrating the superiority of our proposed techniques.