nuScenes
Motional
1,000 annotated driving scenes from Boston and Singapore with a full self-driving sensor suite.
A large public dataset for autonomous driving that captures urban situations with the full sensor suite of a real self-driving car, including cameras, lidar, and radar, with 3D object annotations. A standard benchmark for detection and tracking research.
Access
- Cost
- FreeNo charge to download; hosted in an AWS S3 bucket.
- License
- Provider's own termsnuScenes terms of use (record lists commercial terms link)
- Sign-up
- No sign-up
- How you get it
- API and Bulk downloadJSON, JPEG, PCD
What's in it
- One row is
- One driving scene with sensor sweeps and 3D annotations
- Covers
- Boston and Singapore
- Years
- Not stated
- Updated
- One-time release
- Size
- 1,000 scenes (about 20 seconds each, per the project description)
Quality
One of the most-used autonomous driving benchmarks, with documentation and a devkit. Terms of use restrict some uses.
- Custom terms of use
- Two cities only
Connects to
- nuPlan: Planning benchmark from the same company
- Boreas Autonomous Driving Dataset: All-weather sequences with precise localization ground truth
Questions it could answer
- How well do 3D detectors work at night or in rain?
- How do traffic patterns differ between Boston and Singapore?