Gradient-Free Adversarial Training Against Image Corruption for Learning-based Steering


Abstract

We introduce a simple yet effective framework for improving the robustness of learning algorithm against (input) image corruptions for autonomous driving, due to both internal (e.g., sensor noises and hardware abnormalities) and external factors (e.g., lighting, weather, visibility, and other environmental effects). Using sensitivity analysis with FID-based parameterization, we propose a novel algorithm exploiting basis perturbations to improve the overall performance of autonomous steering and other image processing tasks, such as classification and detection, for self-driving cars. Our model not only improves the performance on the original dataset, but also achieves significant performance improvement on datasets with multiple and unseen perturbations, up to 87% and 77%, respectively. A comparative study drawn between our approach and other SOTA techniques confirms the effectiveness of our technique in improving the robustness of neural network training for learning-based steering and other image processing tasks.

Paper

Gradient-Free Adversarial Training Against Image Corruption for Learning-based Steering
Yu Shen, Laura Zheng, Manli Shu, Weizi Li, Tom Goldstein, Ming C. Lin.

Video

Demo video can be found here.

Slides

Slides can be found here.

Code

The GitHub repository is located here.