Promoting Shape Bias in CNNs: Frequency-Based and Contrastive Regularization for Corruption Robustness
This addresses robustness issues for image classification systems, though it is incremental as it builds on existing regularization techniques.
The paper tackled the problem of CNNs' vulnerability to image corruptions by promoting shape-biased representations, resulting in improved robustness on CIFAR-10-C without degrading clean accuracy.
Convolutional Neural Networks (CNNs) excel at image classification but remain vulnerable to common corruptions that humans handle with ease. A key reason for this fragility is their reliance on local texture cues rather than global object shapes -- a stark contrast to human perception. To address this, we propose two complementary regularization strategies designed to encourage shape-biased representations and enhance robustness. The first introduces an auxiliary loss that enforces feature consistency between original and low-frequency filtered inputs, discouraging dependence on high-frequency textures. The second incorporates supervised contrastive learning to structure the feature space around class-consistent, shape-relevant representations. Evaluated on the CIFAR-10-C benchmark, both methods improve corruption robustness without degrading clean accuracy. Our results suggest that loss-level regularization can effectively steer CNNs toward more shape-aware, resilient representations.