Directly Denoising Diffusion Models

TL;DR

The paper introduces DDDM, a simple, generic diffusion model that enables high-quality image generation with minimal sampling steps, outperforming many existing models without complex samplers or distillation.

Contribution

It proposes a novel directly denoising diffusion approach that simplifies sampling and introduces Pseudo-LPIPS for robust metric evaluation, achieving state-of-the-art results.

Findings

Achieves FID of 2.57 on CIFAR-10 with one-step sampling

Surpasses GANs and distillation models in benchmark performance

Reduces FID to 1.79 with 1000 sampling steps, matching state-of-the-art

Abstract

In this paper, we present the Directly Denoising Diffusion Model (DDDM): a simple and generic approach for generating realistic images with few-step sampling, while multistep sampling is still preserved for better performance. DDDMs require no delicately designed samplers nor distillation on pre-trained distillation models. DDDMs train the diffusion model conditioned on an estimated target that was generated from previous training iterations of its own. To generate images, samples generated from the previous time step are also taken into consideration, guiding the generation process iteratively. We further propose Pseudo-LPIPS, a novel metric loss that is more robust to various values of hyperparameter. Despite its simplicity, the proposed approach can achieve strong performance in benchmark datasets. Our model achieves FID scores of 2.57 and 2.33 on CIFAR-10 in one-step and two-step sampling respectively, surpassing those obtained from GANs and distillation-based models. By extending the sampling to 1000 steps, we further reduce FID score to 1.79, aligning with state-of-the-art methods in the literature. For ImageNet 64x64, our approach stands as a competitive contender against leading models.