AR2: Attention-Guided Repair for the Robustness of CNNs Against Common Corruptions

2025-07-08

AI Summary

Deep neural networks exhibit insufficient robustness against common image degradations—such as noise, blur, and weather-induced distortions—hindering their real-world deployment. To address this, we propose an attention-consistency enhancement method that requires no architectural modification: it leverages Class Activation Maps (CAMs) to align and guide iterative feature-space restoration of degraded images, alternating between attention-guided restoration and standard fine-tuning. Crucially, the method preserves original model accuracy while substantially improving robustness across diverse degradation types. Evaluated on CIFAR-10-C, CIFAR-100-C, and ImageNet-C benchmarks, our approach consistently outperforms existing state-of-the-art methods, particularly under severe degradation conditions. It offers a highly efficient, plug-and-play robustness enhancement paradigm for pre-trained CNNs—requiring neither retraining from scratch nor architectural changes—thus enabling rapid adaptation of off-the-shelf models to challenging real-world visual conditions.

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 learning to rank, online learning, and counterfactual learning for ranking
• Web Mining and Content Analysis: Large pretrained models with web data
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

Deep neural networks suffer from significant performance degradation when exposed to common corruptions such as noise, blur, weather, and digital distortions, limiting their reliability in real-world applications. In this paper, we propose AR2 (Attention-Guided Repair for Robustness), a simple yet effective method to enhance the corruption robustness of pretrained CNNs. AR2 operates by explicitly aligning the class activation maps (CAMs) between clean and corrupted images, encouraging the model to maintain consistent attention even under input perturbations. Our approach follows an iterative repair strategy that alternates between CAM-guided refinement and standard fine-tuning, without requiring architectural changes. Extensive experiments show that AR2 consistently outperforms existing state-of-the-art methods in restoring robustness on standard corruption benchmarks (CIFAR-10-C, CIFAR-100-C and ImageNet-C), achieving a favorable balance between accuracy on clean data and corruption robustness. These results demonstrate that AR2 provides a robust and scalable solution for enhancing model reliability in real-world environments with diverse corruptions.

Problem

• Enhance CNN robustness against common corruptions like noise and blur
• Align class activation maps between clean and corrupted images
• Improve model reliability without architectural changes

Innovation

• Aligns class activation maps for robustness
• Iterative CAM-guided refinement strategy
• No architectural changes required