Baseline Evaluation Methodology for Adversarial Patterns on Object Detection Models

Physically realizable adversarial patterns have be-come a sophisticated form of adversarial ML across industry and academia, with several algorithms that successfully thwart state-of-the-art ML models. However, there are currently no standard practices for evaluating adversarial patterns. We identify two components that provide insight into an adversarial pattern's performance: the seeding of the ...