Alert button

Talk Proposal: Towards the Realistic Evaluation of Evasion Attacks using CARLA

Apr 18, 2019
Cory Cornelius, Shang-Tse Chen, Jason Martin, Duen Horng Chau

Figure 1 for Talk Proposal: Towards the Realistic Evaluation of Evasion Attacks using CARLA
Figure 2 for Talk Proposal: Towards the Realistic Evaluation of Evasion Attacks using CARLA
Figure 3 for Talk Proposal: Towards the Realistic Evaluation of Evasion Attacks using CARLA
Figure 4 for Talk Proposal: Towards the Realistic Evaluation of Evasion Attacks using CARLA

Share this with someone who'll enjoy it:

In this talk we describe our content-preserving attack on object detectors, ShapeShifter, and demonstrate how to evaluate this threat in realistic scenarios. We describe how we use CARLA, a realistic urban driving simulator, to create these scenarios, and how we use ShapeShifter to generate content-preserving attacks against those scenarios.

* Submitted as talk proposal to Dependable and Secure Machine Learning (DSML '19)  
View paper onarxiv icon

Share this with someone who'll enjoy it: