Degradation-Consistent Paired Training for Robust AI-Generated Image Detection

TL;DR

This paper introduces Degradation-Consistent Paired Training (DCPT), a training strategy that explicitly enforces robustness in AI-generated image detection under various corruptions without extra parameters.

Contribution

DCPT explicitly enforces degradation robustness through paired consistency constraints, improving performance without additional model complexity.

Findings

DCPT improves degraded-condition accuracy by 9.1 percentage points.

DCPT enhances JPEG compression robustness by over 15%.

Architectural changes cause overfitting; training objectives are more effective.

Abstract

AI-generated image detectors suffer significant performance degradation under real-world image corruptions such as JPEG compression, Gaussian blur, and resolution downsampling. We observe that state-of-the-art methods, including B-Free, treat degradation robustness as a byproduct of data augmentation rather than an explicit training objective. In this work, we propose Degradation-Consistent Paired Training (DCPT), a simple yet effective training strategy that explicitly enforces robustness through paired consistency constraints. For each training image, we construct a clean view and a degraded view, then impose two constraints: a feature consistency loss that minimizes the cosine distance between clean and degraded representations, and a prediction consistency loss based on symmetric KL divergence that aligns output distributions across views. DCPT adds zero additional parameters and zero inference overhead. Experiments on the Synthbuster benchmark (9 generators, 8 degradation conditions) demonstrate that DCPT improves the degraded-condition average accuracy by 9.1 percentage points compared to an identical baseline without paired training, while sacrificing only 0.9% clean accuracy. The improvement is most pronounced under JPEG compression (+15.7% to +17.9%). Ablation further reveals that adding architectural components leads to overfitting on limited training data, confirming that training objective improvement is more effective than architectural augmentation for degradation robustness.