DRIFT: From Robustness Gaps to Invariance Manifolds for AI-Generated Image Detection

2026-06-05

AI Summary

This work addresses the limited generalization of existing AI-generated image detection methods in open-world scenarios and their lack of adaptive mechanisms. It formulates the task, for the first time, as a structured invariant manifold learning problem under single-class supervision. Building upon a frozen vision foundation model, the authors design a lightweight projection head that decomposes representations into robust and fragile subspaces: the former captures tangential invariance of real images under physically plausible transformations, while the latter retains sensitivity to editing perturbations. Detection is achieved through manifold deviation analysis via multi-scale block-level perturbations, dual-channel invariance signatures, and structured ranking margin optimization, yielding interpretable results. The method significantly outperforms training-free baselines across diverse unseen generators and resolutions and produces invariance-violation heatmaps.

Abstract

The rapid evolution of generative image models challenges existing AI-generated image detectors, particularly in open-world settings with unseen generators. Recent training-free approaches measure robustness gaps in frozen vision foundation models (VFMs), detecting fakes via perturbation-induced embedding drift. However, these methods rely on fixed invariance geometry inherited from pretraining and lack principled adaptation to the detection task. We instead formulate AI-generated image detection as learning a structured invariance manifold of real images under one-class supervision. Building upon a frozen VFM, we introduce lightweight projection heads that decompose representation space into complementary robust and fragile subspaces. The robust subspace is explicitly trained to suppress variations induced by physically plausible imaging transformations, approximating tangent directions of a real-image manifold, while the fragile subspace retains sensitivity to edit-like perturbations. A structured ordering margin enforces hierarchical separation between physical invariance and edit-induced variability, enabling detection as a margin-violation test relative to the learned manifold. At inference, multi-scale patch-wise drift under both transformation families yields a dual-channel invariance signature and interpretable localization. Extensive experiments demonstrate strong open-world generalization across unseen generators and resolutions, consistently outperforming training-free robustness-based baselines while providing interpretable invariance-violation maps.

Problem

• AI-generated image detection
• open-world generalization
• invariance manifold
• robustness gaps
• vision foundation models

Innovation

• invariance manifold
• vision foundation model
• robust subspace
• training-free detection
• open-world generalization