Diffusion Classifier Guidance for Non-robust Classifiers

TL;DR

This paper extends diffusion classifier guidance to non-robust classifiers, addressing stability issues caused by noise sensitivity and proposing methods to improve guidance stability without sacrificing sample quality.

Contribution

It introduces a novel approach for using non-robust classifiers in diffusion guidance, including stabilization techniques inspired by stochastic optimization.

Findings

Non-robust classifiers degrade in accuracy under noisy conditions.

Proposed stabilization methods improve guidance stability.

Sample diversity and quality are maintained with the new approach.

Abstract

Classifier guidance is intended to steer a diffusion process such that a given classifier reliably recognizes the generated data point as a certain class. However, most classifier guidance approaches are restricted to robust classifiers, which were specifically trained on the noise of the diffusion forward process. We extend classifier guidance to work with general, non-robust, classifiers that were trained without noise. We analyze the sensitivity of both non-robust and robust classifiers to noise of the diffusion process on the standard CelebA data set, the specialized SportBalls data set and the high-dimensional real-world CelebA-HQ data set. Our findings reveal that non-robust classifiers exhibit significant accuracy degradation under noisy conditions, leading to unstable guidance gradients. To mitigate these issues, we propose a method that utilizes one-step denoised image predictions and implements stabilization techniques inspired by stochastic optimization methods, such as exponential moving averages. Experimental results demonstrate that our approach improves the stability of classifier guidance while maintaining sample diversity and visual quality. This work contributes to advancing conditional sampling techniques in generative models, enabling a broader range of classifiers to be used as guidance classifiers.