1 code implementation • 3 Jun 2022 • Zhengyu Zhao, Nga Dang, Martha Larson
In this paper, we propose that adversarial images should be evaluated based on semantic mismatch, rather than label mismatch, as used in current work.
