Delving Deep into the Generalization of Vision Transformers under Distribution Shifts

• Published in: arXiv.org
• Main Authors: Zhang, Chongzhi, Zhang, Mingyuan, Zhang, Shanghang, Jin, Daisheng, Zhou, Qiang, Cai, Zhongang, Zhao, Haiyu, Liu, Xianglong, Liu, Ziwei
• Format: Paper
• Language: English
• Published: Ithaca Cornell University Library, arXiv.org 08.03.2022
• Subjects: Information theory; Parameter sensitivity; Supervised learning; Taxonomy; Training

Vision Transformers (ViTs) have achieved impressive performance on various vision tasks, yet their generalization under distribution shifts (DS) is rarely understood. In this work, we comprehensively study the out-of-distribution (OOD) generalization of ViTs. For systematic investigation, we first p...