Robust Adversarial Perturbation on Deep Proposal-based Models

TL;DR

This paper introduces a robust adversarial perturbation method targeting the Region Proposal Network in deep object detectors and segmentation models, effectively degrading their performance in black-box settings.

Contribution

It proposes a novel loss function combining label and shape losses, optimized via a gradient-based iterative algorithm for universal attacks.

Findings

Successfully attacks 6 state-of-the-art detectors

Degrades performance in black-box scenarios

Effective on MS COCO 2014 dataset

Abstract

Adversarial noises are useful tools to probe the weakness of deep learning based computer vision algorithms. In this paper, we describe a robust adversarial perturbation (R-AP) method to attack deep proposal-based object detectors and instance segmentation algorithms. Our method focuses on attacking the common component in these algorithms, namely Region Proposal Network (RPN), to universally degrade their performance in a black-box fashion. To do so, we design a loss function that combines a label loss and a novel shape loss, and optimize it with respect to image using a gradient based iterative algorithm. Evaluations are performed on the MS COCO 2014 dataset for the adversarial attacking of 6 state-of-the-art object detectors and 2 instance segmentation algorithms. Experimental results demonstrate the efficacy of the proposed method.