Clean-Annotation Backdoor Attack against Lane Detection Systems in the Wild

Abstract

We present the first backdoor attack against the lane detection systems in the physical world. Modern autonomous vehicles adopt various deep learning methods to train lane detection models, making it challenging to devise a universal backdoor attack technique. In our solution, (1) we propose a novel semantic trigger design, which leverages the traffic cones with specific poses and locations to activate the backdoor. Such trigger can be easily realized under the physical setting, and looks very natural not to be detected. ( 2) We introduce a new cleanannotation approach to generate poisoned samples. These samples have correct annotations but are still capable of embedding the backdoor to the model. Comprehensive evaluations on public datasets and physical autonomous vehicles demonstrate that our backdoor attack is effective, stealthy and robust.

DBA: downsampling-based adversarial attack in medical image analysis

“…However, Han et.al. [24,33] found that this process triggers new adversarial attacks: an adversary can unknowingly modify the original image, and after scaling down, the original image becomes the desired adversarial image. Inspired by these works, we propose an adversarial attack method for medical image segmentation, which is a generic attack method that can be independent of specific machine learning models.…”