Transferable Adversarial Examples for Anchor Free Object Detection
This addresses a security vulnerability in object detection systems, particularly for anchor-free models, with incremental novelty as it extends existing adversarial attack research to a new detector type.
The paper tackles the problem of adversarial attacks on anchor-free object detectors, presenting the first such attack that uses category-wise attacks and high-level semantic information to generate transferable adversarial examples, achieving state-of-the-art performance and transferability across datasets.
Deep neural networks have been demonstrated to be vulnerable to adversarial attacks: subtle perturbation can completely change prediction result. The vulnerability has led to a surge of research in this direction, including adversarial attacks on object detection networks. However, previous studies are dedicated to attacking anchor-based object detectors. In this paper, we present the first adversarial attack on anchor-free object detectors. It conducts category-wise, instead of previously instance-wise, attacks on object detectors, and leverages high-level semantic information to efficiently generate transferable adversarial examples, which can also be transferred to attack other object detectors, even anchor-based detectors such as Faster R-CNN. Experimental results on two benchmark datasets demonstrate that our proposed method achieves state-of-the-art performance and transferability.