D²-City: Large-Scale Dashcam Video Dataset of Diverse Traffic Scenarios

China is the geographic scope for this dataset of more than 10,000 dashcam videos recorded in 720p HD or 1080p FHD. It was created by Zhengping Che and is hosted on DiDi GAIA's platform. Around 1,000 videos have per-frame detection and tracking annotations for road objects, with additional videos annotated in keyframes.

Use Cases

• Training object detection models based on bounding box annotations for cars, pedestrians, and other road users.
• Developing multi-object tracking algorithms based on the provided tracking IDs across video frames.
• Testing model robustness based on the dataset's diversity in weather, road, and traffic conditions across several Chinese cities.

Strengths

• More than 10,000 dashcam videos provide a substantial volume of raw visual data.
• Around 1,000 videos include detailed per-frame detection and tracking annotations for 11 object categories.
• Data diversity is noted, with collection from several cities and varying weather, road, and traffic conditions.

Limitations

• License is closed, which restricts commercial use and redistribution.
• Column-level documentation is absent; field semantics must be inferred after download.
• Last update date is unknown; freshness unverified.

Provenance

Source

DiDi GAIA's platform

Collection Method

Dashcam video recordings from several cities in China.

Geography

Several cities throughout China