What autonomous driving dataset formats are available?

LabelSets supports KITTI 3D format with 3D bounding boxes and calibration files, as well as COCO JSON for camera-only datasets, and Parquet for large-scale tabular sensor data.

What types of sensors are covered?

Camera (monocular and stereo), LiDAR point clouds (.pcd, .bin), radar, and multi-modal sensor fusion datasets are all available. Many datasets include synchronized sensor streams.

Dataset Category

Autonomous Vehicle & Self-Driving Datasets

Q: Can I sell proprietary driving data collected from my fleet?

Yes. If you have the rights to your driving data and it has been properly anonymized (faces and license plates blurred), you can list it on LabelSets and earn 85% of each sale.

Camera, LiDAR, and sensor fusion datasets for perception, prediction, and planning in autonomous systems. KITTI-compatible, quality-verified.

Browse AV Datasets → Sell Your Dataset

Autonomous Driving Tasks Covered

Every perception and prediction task needed to train a modern autonomous driving stack.

📡

LiDAR Point Clouds

3D point cloud datasets with 3D bounding box annotations for vehicle, pedestrian, and cyclist detection at range.

🎥

Camera Perception

Multi-camera surround-view datasets with 2D/3D object detection, lane segmentation, and depth estimation labels.

🛣️

Lane & Road Detection

Lane marking segmentation, road surface detection, and HD map annotation datasets for path planning.

🚦

Traffic Sign & Signal

Traffic sign classification and traffic light state detection datasets across multiple countries and lighting conditions.

🌐

Sensor Fusion

Synchronized LiDAR + camera + radar sequences with calibration matrices and temporal correspondences.

🚶

Pedestrian & Behavior

Pedestrian trajectory, intent prediction, and interaction datasets for social force and transformer-based models.

Frequently Asked Questions

KITTI 3D format (with 3D bounding boxes and camera calibration files) is the primary supported format. COCO JSON is supported for camera-only detection, and Parquet is available for large-scale tabular sensor logs.

Camera (monocular and stereo), LiDAR point clouds (.pcd, .bin), radar, and multi-modal sensor fusion datasets. Many include synchronized sensor streams with timestamp alignment.

Yes, if you hold the rights to the data and it has been properly anonymized (faces and license plates blurred). Upload, pass verification, set your price, and earn 85% of every sale.

Yes. Sellers can tag datasets by weather condition (rain, fog, night, snow) and scenario type (intersection, highway, construction zone) to help buyers find safety-critical edge cases.