Towards Compositional Adversarial Robustness: Generalizing Adversarial Training to Composite Semantic PerturbationsLei HsiungYun-Yun Tsaiet al.2023CVPR 2023