FAIR: Feature-Augmented Implicit Regularization for AI-generated Fake Image Detection

2026-07-24

AI Summary

Existing methods for detecting AI-generated images suffer from limited generalization and a tendency to overfit to low-level textural features in training data. To address this, this work proposes Feature-Augmented Implicit Regularization (FAIR), which introduces Scene Composition Structure (SCS) as an orthogonal macroscopic prior during training to impose geometric constraints on the model’s optimization trajectory. This suppresses shortcut learning based on textures and enables domain-invariant learning without modifying the model architecture. Notably, the SCS prior is entirely discarded during inference, incurring no additional computational overhead. Experiments demonstrate that FAIR significantly improves cross-generator detection performance across five large-scale benchmarks, achieving up to an 8.04% absolute gain in accuracy, and attains state-of-the-art robustness in zero-shot transfer scenarios.

Abstract

Generalization remains a critical bottleneck in AI-generated image detection. Because many modern generators are proprietary or adversarially modified, existing detectors overfit to the low-level textural patterns of accessible training data, resulting in severe failures on unseen domains. Conventional regularization techniques (e.g., $L_1$/$L_2$ norms, Dropout) apply indiscriminate parametric constraints and fail to provide the domain-invariant structure necessary for cross-generator robustness. To address this, we propose Feature-Augmented Implicit Regularization (FAIR). FAIR introduces an orthogonal, macro-structural prior, specifically, Scene Composition Structure (SCS), during training to geometrically constrain the model's optimization trajectory. By augmenting the primary feature space with domain-invariant SCS features, FAIR explicitly penalizes texture-biased shortcut learning. Crucially, this structural prior is entirely discarded at inference, yielding a smoothed, generalized decision boundary with zero architectural or computational overhead. Extensive evaluations across five massive benchmarks demonstrate that integrating FAIR into state-of-the-art detectors significantly improves cross-generator generalization, boosting accuracy by up to 8.04% and establishing new state-of-the-art robustness in zero-shot transfer scenarios.

Problem

• generalization
• AI-generated image detection
• domain-invariant
• texture bias
• cross-generator robustness

Innovation

• Feature-Augmented Implicit Regularization
• Scene Composition Structure
• domain-invariant representation
• cross-generator generalization
• shortcut learning suppression