nuScenes Mini Autonomous Driving Dataset

nuScenes Mini is a subset of the full nuScenes dataset for autonomous vehicle perception. It contains annotated sensor data from a vehicle equipped with cameras, LIDAR, and radar. The dataset was created by Motional (formerly nuTonomy) for research in autonomous driving.

Use Cases

• Train 3D object detection models using LIDAR point cloud data and camera images.
• Develop sensor fusion algorithms by aligning data from cameras, LIDAR, and radar sensors.
• Benchmark tracking and prediction models using annotated object trajectories across 20-second scenes.
• Validate perception systems against human-annotated 3D bounding boxes for 23 object classes.

Strengths

• Contains synchronized data from 6 cameras, 1 LIDAR, and 5 radar sensors.
• Provides 3D bounding box annotations for 23 object classes across multiple scenes.

Limitations

• The 'Mini' version is a small subset, significantly limiting the scale for training large models.
• Annotations are limited to 20-second scene snippets, which may not capture long-term behavioral patterns.

Provenance

Source

Motional (formerly nuTonomy)

Collection Method

Data collected from an autonomous vehicle platform driving in Boston and Singapore.

Geography

Boston, USA and Singapore