Generated Images Are Easier to Forget: A Machine Unlearning Perspective for Synthetic Image Detection

2026-08-01

AI Summary

This work addresses the limitations of existing generated image detection methods, which rely on manually annotated data and exhibit constrained generalization. It reframes the problem as a machine unlearning task, revealing that generated images undergo faster feature degradation during model unlearning compared to real images. Building on this insight, the paper proposes a novel, annotation-free detection paradigm: it leverages parameter pruning to achieve data-free unlearning and introduces a targeted unlearning strategy for data-driven optimization. Extensive experiments demonstrate that the proposed approach significantly outperforms current state-of-the-art methods across multiple benchmarks, thereby validating the effectiveness and novelty of leveraging unlearning dynamics for detecting synthetic imagery.

Abstract

Robust detection of generated images is critical to counter the misuse of generative models. Existing methods primarily depend on learning from human-annotated training datasets, limiting their generalization to unseen distributions. In contrast, large-scale vision models (LVMs) pre-trained on web-scale datasets exhibit exceptional generalization power through exposure to diverse distributions, offering a transformative paradigm for this task. However, our experimental results reveal that LVMs pre-trained on natural-image-dominated data can effectively capture the features of both natural and generated images, yielding comparably low losses and thus limited discriminative capacity between them. This prompts a key question: When and how do LVMs exhibit different behaviors when capturing features of natural and generated images? This investigation reveals an insight: during unlearning, LVMs exhibit disparate forgetting dynamics with feature degradation for generated images escalating faster than natural ones. Inspired by the disparate dynamics, we introduce two detection methods: 1) data-free detection, which prunes model parameters to induce unlearning without data access, and 2) data-driven detection, which optimizes LVMs to unlearn knowledge tied to generated images. Extensive experiments conducted on various benchmarks demonstrate that our unlearning-based approach outperforms conventional detection methods. By recasting the detection task as a problem of machine unlearning, our work establishes a new paradigm for generated image detection.

Problem

• synthetic image detection
• machine unlearning
• large-scale vision models
• generated images
• feature forgetting

Innovation

• machine unlearning
• synthetic image detection
• large-scale vision models
• forgetting dynamics
• data-free detection