Epistemic Uncertainty for Generated Image Detection

TL;DR

This paper presents a new method for detecting AI-generated images by leveraging epistemic uncertainty, which increases for generated images due to distributional shifts, using pre-trained vision models to estimate uncertainty.

Contribution

The paper introduces a novel approach that uses epistemic uncertainty estimation with pre-trained models to effectively detect AI-generated images, addressing security concerns.

Findings

High accuracy in distinguishing generated images from natural ones.

Effective use of pre-trained vision models for uncertainty estimation.

Robustness across various generative models and datasets.

Abstract

We introduce a novel framework for AI-generated image detection through epistemic uncertainty, aiming to address critical security concerns in the era of generative models. Our key insight stems from the observation that distributional discrepancies between training and testing data manifest distinctively in the epistemic uncertainty space of machine learning models. In this context, the distribution shift between natural and generated images leads to elevated epistemic uncertainty in models trained on natural images when evaluating generated ones. Hence, we exploit this phenomenon by using epistemic uncertainty as a proxy for detecting generated images. This converts the challenge of generated image detection into the problem of uncertainty estimation, underscoring the generalization performance of the model used for uncertainty estimation. Fortunately, advanced large-scale vision models pre-trained on extensive natural images have shown excellent generalization performance for various scenarios. Thus, we utilize these pre-trained models to estimate the epistemic uncertainty of images and flag those with high uncertainty as generated. Extensive experiments demonstrate the efficacy of our method. Code is available at https://github.com/tmlr-group/WePe.