Robustifying Deep Vision Models Through Shape Sensitization
This addresses the texture bias issue in deep neural networks for computer vision, offering a lightweight method to enhance model robustness, though it is incremental as it builds on existing augmentation strategies.
The paper tackles the problem of deep vision models' over-reliance on texture features by proposing an adversarial augmentation technique that incentivizes learning holistic shapes, resulting in significant improvements in classification accuracy and robustness, such as up to 6% absolute gains on classification accuracy and up to 28% gains on natural adversarial datasets.
Recent work has shown that deep vision models tend to be overly dependent on low-level or "texture" features, leading to poor generalization. Various data augmentation strategies have been proposed to overcome this so-called texture bias in DNNs. We propose a simple, lightweight adversarial augmentation technique that explicitly incentivizes the network to learn holistic shapes for accurate prediction in an object classification setting. Our augmentations superpose edgemaps from one image onto another image with shuffled patches, using a randomly determined mixing proportion, with the image label of the edgemap image. To classify these augmented images, the model needs to not only detect and focus on edges but distinguish between relevant and spurious edges. We show that our augmentations significantly improve classification accuracy and robustness measures on a range of datasets and neural architectures. As an example, for ViT-S, We obtain absolute gains on classification accuracy gains up to 6%. We also obtain gains of up to 28% and 8.5% on natural adversarial and out-of-distribution datasets like ImageNet-A (for ViT-B) and ImageNet-R (for ViT-S), respectively. Analysis using a range of probe datasets shows substantially increased shape sensitivity in our trained models, explaining the observed improvement in robustness and classification accuracy.