Score Mismatching for Generative Modeling

TL;DR

This paper introduces a novel score-based generative model that enables one-step sampling by training a generator with a score network, reducing computational costs and outperforming existing models on benchmark datasets.

Contribution

The authors propose a standalone generator trained with a score network to achieve one-step sampling, eliminating iterative processes in score-based generative modeling.

Findings

Outperforms Consistency Model and Denoising Score Matching on CIFAR-10

Requires only 10 diffusion steps for training

Enables one-step image generation during sampling

Abstract

We propose a new score-based model with one-step sampling. Previously, score-based models were burdened with heavy computations due to iterative sampling. For substituting the iterative process, we train a standalone generator to compress all the time steps with the gradient backpropagated from the score network. In order to produce meaningful gradients for the generator, the score network is trained to simultaneously match the real data distribution and mismatch the fake data distribution. This model has the following advantages: 1) For sampling, it generates a fake image with only one step forward. 2) For training, it only needs 10 diffusion steps.3) Compared with consistency model, it is free of the ill-posed problem caused by consistency loss. On the popular CIFAR-10 dataset, our model outperforms Consistency Model and Denoising Score Matching, which demonstrates the potential of the framework. We further provide more examples on the MINIST and LSUN datasets. The code is available on GitHub.