References & Citations
Computer Science > Computer Vision and Pattern Recognition
Title: Learning Coated Adversarial Camouflages for Object Detectors
(Submitted on 1 Sep 2021 (v1), last revised 14 May 2022 (this version, v3))
Abstract: An adversary can fool deep neural network object detectors by generating adversarial noises. Most of the existing works focus on learning local visible noises in an adversarial "patch" fashion. However, the 2D patch attached to a 3D object tends to suffer from an inevitable reduction in attack performance as the viewpoint changes. To remedy this issue, this work proposes the Coated Adversarial Camouflage (CAC) to attack the detectors in arbitrary viewpoints. Unlike the patch trained in the 2D space, our camouflage generated by a conceptually different training framework consists of 3D rendering and dense proposals attack. Specifically, we make the camouflage perform 3D spatial transformations according to the pose changes of the object. Based on the multi-view rendering results, the top-n proposals of the region proposal network are fixed, and all the classifications in the fixed dense proposals are attacked simultaneously to output errors. In addition, we build a virtual 3D scene to fairly and reproducibly evaluate different attacks. Extensive experiments demonstrate the superiority of CAC over the existing attacks, and it shows impressive performance both in the virtual scene and the real world. This poses a potential threat to the security-critical computer vision systems.
Submission history
From: Zhisong Pan [view email][v1] Wed, 1 Sep 2021 00:18:17 GMT (20366kb,D)
[v2] Thu, 18 Nov 2021 07:56:06 GMT (35669kb,D)
[v3] Sat, 14 May 2022 12:50:40 GMT (35232kb,D)
Link back to: arXiv, form interface, contact.