Reproduce “Understanding and Improving Robustness of Vision Transformers through Patch-based Negative Augmentation” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Yao Qin, Chiyuan Zhang, Ting Chen, Balaji Lakshminarayanan, Alex Beutel, Xuezhi Wang

The authors introduce patch-based negative augmentation to improve Vision Transformer robustness to distribution shift. The method pairs labeled images with patch-shuffled, rotated, or infilled views, penalizing class-preserving predictions on those views while retaining ordinary positive augmentation. With AugMix, ViT-B/16 reaches 26.0% on ImageNet-R versus 24.9%, and 79.0% versus 78.8% on ImageNet-1k; gains are not universal.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.