Weakly supervised method improves detection of ai-edited image regions
Less Supervision, Better Generalization: Weakly Supervised Fake Region Localization in Diffusion-Edited Images
Computer Vision and Pattern Recognition
Summary
Identifying which parts of an image were altered by AI is important but hard because the changes are often subtle and spread out. The authors designed a new method that doesn't need detailed labels showing exactly which pixels were edited. Instead, their approach learns from just knowing if an image is real or fake and uses differences in image reconstruction to find altered regions. This method generalizes better than older ones that required detailed pixel-level labels.
What this means in practice
- •For digital forensics teams: Detect altered regions in AI-edited images without needing detailed annotations, improving forensic analysis of manipulated media.
- •For content moderation teams: Support detection of partially AI-altered images at scale using only image-level real/fake labels, aiding in automated content verification.
Authors
Junhee Lee, Donghyeon Jeon, Taeoh Kim, Beomyoung Kim, MyeongAh Cho
Abstract
Localizing AI-edited regions is essential for interpretable forensic analysis, but remains challenging due to subtle and spatially distributed artifacts that are misaligned with semantic or object boundaries. Existing approaches rely on pixel-level supervision from controlled editing pipelines, which is difficult to scale and can introduce misleading signals: artifacts frequently extend beyond annotated regions, while out-of-mask pixels are treated as authentic. This limits models' ability to capture transferable evidence and generalize across generators and datasets. To address these issues, we propose ReGFLoW, a Reconstruction-Guided Fake Localization framework under Weak supervision, which is the first weakly supervised approach for diffusion-edited fake region localization. ReGFLoW requires only real/fake labels at the image-level and uses diffusion reconstruction errors as dense spatial guidance to inject them into both feature and score spaces. Furthermore, by artifact-centric multiple instance learning, ReGFLoW utilizes localized diffusion evidence without relying on semantic-affinity or boundary-based pseudo-mask priors. Extensive experiments demonstrate that ReGFLoW achieves stronger out-of-domain generalization than fully supervised learning baselines.