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

2026-04-11

AI Summary

Current AI-generated image detectors suffer significant performance degradation under realistic image degradations such as JPEG compression and Gaussian blur. This work proposes a parameter- and overhead-free paired training strategy that explicitly optimizes for degradation robustness for the first time. By constructing clean-degraded image pairs and jointly enforcing consistency constraints at both the feature level (via cosine distance) and the prediction level (via symmetric KL divergence), the method enhances model generalization. Evaluated on the SynthBench benchmark, the approach improves average detection accuracy by 9.1 percentage points, with gains of 15.7%–17.9% specifically under JPEG compression, while incurring only a marginal 0.9% drop in accuracy on pristine images.

Technology Category

• Computer Vision: Adversarial Attacks & Robustness
• Machine Learning: Adversarial Learning & Robustness
• Natural Language Processing: Safety and Robustness

Application Category

• Search and Retrieval-Augmented AI: Web evaluation methodologies and metrics
• Web Mining and Content Analysis: Robustness and generalizability of Web mining methods
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

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.

Problem

• AI-generated image detection
• image degradation
• robustness
• JPEG compression
• Gaussian blur

Innovation

• Degradation-Consistent Paired Training
• AI-generated image detection
• robustness to image corruption
• feature consistency loss
• prediction consistency loss