PBCAT: Patch-based composite adversarial training against physically realizable attacks on object detection

2025-06-30

AI Summary

Physical adversarial attacks—such as adversarial patches and textures—pose significant threats to object detectors, yet existing robust training methods lack a unified framework capable of defending against diverse, real-world physical attack modalities. Method: This paper proposes a patch-based composite adversarial training framework that jointly optimizes gradient-guided local adversarial patches and globally imperceptible perturbations, integrated with multi-scale data augmentation and end-to-end joint training. Contribution/Results: To the best of our knowledge, this is the first approach enabling unified modeling and defense against multiple physical attack types. It exhibits strong generalization, effectively mitigating both seen and unseen attacks. Under state-of-the-art adversarial texture attacks, it improves detection accuracy by 29.7% over prior methods, substantially enhancing detector security and robustness in practical deployment scenarios.

Technology Category

• Computer Vision: Adversarial Attacks & Robustness
• Machine Learning: Adversarial Learning & Robustness
• Search and Optimization: Adversarial Search

Application Category

• User Modeling, Personalization and Recommendation: Attacks and countermeasures in recommendation systems
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Security and Privacy: Large-scale security measurements

Abstract

Object detection plays a crucial role in many security-sensitive applications. However, several recent studies have shown that object detectors can be easily fooled by physically realizable attacks, eg, adversarial patches and recent adversarial textures, which pose realistic and urgent threats. Adversarial Training (AT) has been recognized as the most effective defense against adversarial attacks. While AT has been extensively studied in the $l_infty$ attack settings on classification models, AT against physically realizable attacks on object detectors has received limited exploration. Early attempts are only performed to defend against adversarial patches, leaving AT against a wider range of physically realizable attacks under-explored. In this work, we consider defending against various physically realizable attacks with a unified AT method. We propose PBCAT, a novel Patch-Based Composite Adversarial Training strategy. PBCAT optimizes the model by incorporating the combination of small-area gradient-guided adversarial patches and imperceptible global adversarial perturbations covering the entire image. With these designs, PBCAT has the potential to defend against not only adversarial patches but also unseen physically realizable attacks such as adversarial textures. Extensive experiments in multiple settings demonstrated that PBCAT significantly improved robustness against various physically realizable attacks over state-of-the-art defense methods. Notably, it improved the detection accuracy by 29.7% over previous defense methods under one recent adversarial texture attack.

Problem

• Defends object detectors against physically realizable attacks
• Unifies adversarial training for patches and global perturbations
• Improves robustness against unseen adversarial textures

Innovation

• Unified adversarial training for diverse physical attacks
• Combines gradient-guided patches and global perturbations
• Enhances robustness against unseen adversarial textures