nuCarla: A Large-Scale Synthetic Camera Dataset for Autonomous Driving Perception

A dataset of 1,000 driving scenarios, split into 700 training, 150 validation, and 150 testing examples, matching the nuScenes split. nuCarla is a large-scale, camera-based perception dataset developed in the CARLA simulator by zhijieq, last updated in January 2026. It is designed to facilitate training of effective perception representations for end-to-end autonomous driving development.

Use Cases

• Training object detection models based on the 6 object classes mentioned in the description
• Evaluating perception robustness across the 14 simulated weather conditions
• Developing end-to-end driving models using the nuScenes-compatible, camera-based data
• Benchmarking model performance across the 9 distinct simulated maps

Strengths

• Contains 1,000 total scenarios with a defined 700/150/150 train/validation/test split
• Features 9 distinct simulated maps and 14 weather conditions for environmental diversity
• Designed to be compatible with the established nuScenes dataset format

Limitations

• Column-level documentation is absent; field semantics must be inferred after download
• Row count is unknown, which may limit suitability assessment
• Data is synthetic, generated in the CARLA simulator, which may not fully reflect real-world conditions

Provenance

Source

Generated in the CARLA simulator by author zhijieq.

Collection Method

Synthetically generated within a driving simulation environment.

Time Range

null

Freshness

Last updated 2026-01-02 20:00:06; freshness should be verified

Geography

null