PGC: Peak-Guided Calibration for Generalizable AI-Generated Image Detection

2026-05-20

AI Summary

Existing methods for detecting AI-generated images rely on global representations, making them susceptible to high-fidelity content and less effective at capturing subtle forgery artifacts. To address this limitation, this work proposes the Peak-Guided Calibration (PGC) framework, which employs a peak-focusing mechanism to aggregate the most discriminative local features and calibrate global decision-making, thereby enhancing sensitivity to faint generative artifacts. The approach innovatively introduces a peak-sensitive aggregation strategy that strengthens the guidance of critical local cues in shaping the global judgment. Additionally, the authors construct CommGen15, a challenging benchmark comprising samples from 15 commercial generative models. Experimental results demonstrate that PGC achieves state-of-the-art performance, improving average accuracy by 12.3% on CommGen15 and yielding gains of 2.1%, 3.5%, and 1.7% on GenImage, AIGI, and UniversalFakeDetect benchmarks, respectively.

Abstract

The rapid evolution of generative AI, from GANs to modern diffusion models, has resulted in increasingly subtle discriminative clues. These fine-grained signals are often overshadowed by dominant, high-fidelity image content (e.g., the main subject), limiting the reliability of existing detectors that predominantly rely on global representations. To address this challenge, we propose the Peak-Guided Calibration (PGC) framework. PGC introduces a novel strategy that aggregates salient features via a peak-focusing mechanism. Specifically, by employing a peak-sensitive aggregation that accentuates the most discriminative local clues, PGC leverages these critical signals to calibrate the global decision. This approach recovers subtle patterns that would otherwise be submerged in the global context. Furthermore, to better simulate real-world threats, we introduce the CommGen15 dataset, a challenging benchmark comprising samples from 15 commercial models. Extensive experiments demonstrate that PGC achieves state-of-the-art performance. Specifically, it improves mean accuracy by +12.3% on our CommGen15 dataset, and sets new records on standard benchmarks, including GenImage (+2.1%), AIGI (+3.5%), and UniversalFakeDetect (+1.7%). Code is available at https://github.com/xiaoyu6868/PGC.

Problem

• AI-generated image detection
• discriminative clues
• global representation
• subtle patterns
• generalizable detection

Innovation

• Peak-Guided Calibration
• AI-generated image detection
• local discriminative clues
• feature aggregation
• generalizable detection