CLUE-AD: A Context-Based Method for Labeling Unobserved Entities in Autonomous Driving Data

Generating high-quality annotations for object detection and recognition is a challenging important task, especially in relation to safety-critical applications such as autonomous driving (AD). Due the difficulty of perception situations occlusion, degraded weather, sensor failure, objects can go unobserved unlabeled. In this paper, we present CLUE-AD, general-purpose method detecting labeling entities by leveraging continuity assumption within context scene. This dataset-agnostic, supporting any existing future AD datasets. Using real-world dataset representing complex urban scenes, demonstrate applicability CLUE-AD augmenting scene data with new labels.