Boosting the transferability of adversarial attacks with global momentum initialization

Jiafeng Wang, Zhaoyu Chen, Kaixun Jiang, Dingkang Yang, Lingyi Hong, Pinxue Guo, Haijing Guo, Wenqiang Zhang

Abstract: Highlights•We analyze gradient consistency in the context of adversarial attacks for the first time.•Based on gradient consistency, we identify the gradient elimination issue.•We propose global momentum initialization to improve the transferability.•We verify the performance of the proposed method in both image and video domains.