Comma1M: One-Minute Driving Segments with Video and Localization

One million minutes of driving data collected from real user vehicles equipped with comma devices. The dataset includes road-camera video and full localization data for each segment. It was created by commaai and spans several hardware generations, including comma two, comma three, comma 3X, and comma four.

Use Cases

• Train computer vision models for road scene understanding based on the video segments.
• Develop and validate sensor fusion algorithms using the synchronized video and localization data.
• Benchmark end-to-end driving models on real-world, multi-generational hardware data.
• Study the impact of varying camera specifications (resolution, field of view) on model performance.

Strengths

• Contains one million minutes of real-world driving data.
• Includes synchronized video and full localization data for each segment.
• Data spans multiple hardware generations (comma two, three, 3X, and four), providing device diversity.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Camera resolution and field of view are not uniform across the dataset due to varying hardware.
• Row count is unknown, which may limit suitability assessment.

Provenance

Source

commaai

Collection Method

Recorded by comma devices installed in real user vehicles.

Freshness

Last updated 2026-07-14 22:53:06; freshness should be verified.