Scalability in Perception for Autonomous Driving: Waymo Open Dataset

This paper introduces the Waymo Open Dataset, a large-scale benchmark meant to align autonomous-driving research with real-world conditions. It comprises 1150 scenes, each spanning 20 seconds, of well-synchronized and calibrated high-quality LiDAR and camera data captured across a range of urban and suburban geographies. The data is exhaustively annotated with 2D camera-image and 3D LiDAR bounding boxes that maintain consistent identifiers across frames, and the authors supply strong baselines for both 2D and 3D detection and tracking.

The authors report that, by their proposed diversity metric, the dataset is 15 times more diverse than the largest previously available camera-plus-LiDAR dataset. They further study how dataset size and geographic generalization affect 3D detection methods. By offering scale, quality, and diversity, the dataset was released to help the research community address the generalization challenges the authors describe as central to the viability of self-driving technology.