Improving Diffusion-Based Generative Models via Approximated Optimal Transport

TL;DR

This paper presents Approximated Optimal Transport (AOT), a new training scheme for diffusion models that improves image quality and reduces sampling steps by integrating optimal transport principles, achieving state-of-the-art results.

Contribution

The paper introduces AOT, a novel training method that incorporates optimal transport into diffusion models, leading to better denoiser estimation and sampling efficiency.

Findings

Achieved low FID scores with fewer sampling steps.

Enhanced diffusion models' denoiser accuracy and sampling quality.

Set new state-of-the-art FID scores with AOT-guided training.

Abstract

We introduce the Approximated Optimal Transport (AOT) technique, a novel training scheme for diffusion-based generative models. Our approach aims to approximate and integrate optimal transport into the training process, significantly enhancing the ability of diffusion models to estimate the denoiser outputs accurately. This improvement leads to ODE trajectories of diffusion models with lower curvature and reduced truncation errors during sampling. We achieve superior image quality and reduced sampling steps by employing AOT in training. Specifically, we achieve FID scores of 1.88 with just 27 NFEs and 1.73 with 29 NFEs in unconditional and conditional generations, respectively. Furthermore, when applying AOT to train the discriminator for guidance, we establish new state-of-the-art FID scores of 1.68 and 1.58 for unconditional and conditional generations, respectively, each with 29 NFEs. This outcome demonstrates the effectiveness of AOT in enhancing the performance of diffusion models.