Best Paper Award

VATLD: A Visual Analytics System to Assess, Understand and Improve Traffic Light Detection

Liang Gou, Lincan Zou, Nanxiang Li, Michael Hofmann, Shekar Arvind Kumar, Axel Wendt, Liu Ren

View presentation: 2020-10-27T15:06:00Z GMT-0600 Change your timezone on the schedule page
2020-10-27T15:06:00Z
Exemplar figure
A visual analytics system, VATLD, is presented to assess, understand, and improve the accuracy and robustness of traffic light detectors in autonomous driving applications. The VATLD user interface: (a) Summary and navigation of key performance statistics; (b) Visual landscapes of traffic lights (upper) and performance scores (lower) over the first two PCA components of semantic dimensions; (c) A live view to detect a traffic light from either real data or adversarial; (d) Ranked latent dimensions with information of semantics, performance and gradients.
Keywords

Traffic light detection, representation learning, semantic adversarial learning, model diagnosing, autonomous driving

Abstract

Traffic light detection is crucial for environment perception and decision-making in autonomous driving. State-of-the-art detectors are built upon deep Convolutional Neural Networks (CNN) and have exhibited promising performance. However, one looming concern with CNN based detectors is how to thoroughly evaluate the performance of accuracy and robustness before they can be deployed to autonomous vehicles. In this work, we propose a visual analytics system, VATLD, equipped with a disentangled representation learning and semantic adversarial learning, to assess, understand, and improve the accuracy and robustness of traffic light detectors in autonomous driving applications. The disentangled representation learning extracts data semantics to augment human cognition with human-friendly visual summarization, and the semantic adversarial learning efficiently exposes the interpretable robustness risks and enables minimal human interaction for actionable insights. We also demonstrate the effectiveness of various performance improvement strategies derived from actionable insights with our visual analytics system, VATLD, and illustrate the practical implications for safety-critical applications in autonomous driving.