Awesome 3D LiDAR Datasets

Overview

Awesome 3D LiDAR Datasets collects publicly available LiDAR datasets for research in autonomous driving, robotics, and 3D scene understanding.

Autonomous Driving Datasets

KITTI

Pioneer autonomous driving dataset:

• Velodyne 64-line LiDAR
• Camera images
• GPS/IMU data
• 3D object annotations
• Multiple tasks (detection, tracking, segmentation)

nuScenes

Large-scale autonomous driving:

• 1000 scenes, 20 seconds each
• Full 360-degree coverage
• 6 cameras
• 5 radars
• 1 LiDAR
• 3D bounding boxes
• Tracking IDs

Waymo Open Dataset

Large and diverse:

• 1,000 segments of driving data
• High-resolution LiDAR
• 5 LiDAR sensors
• Multiple cameras
• 12M 3D labels

Argoverse

Two versions for different tasks:

• Motion forecasting
• 3D tracking
• HD maps
• Multiple cities

Urban/Outdoor Datasets

SemanticKITTI

Semantic segmentation:

• Built on KITTI
• Point-wise labels
• 28 classes
• Sequential data

Toronto-3D

Urban outdoor dataset:

• Mobile LiDAR
• Point-wise labels
• 8 object classes
• Large-scale coverage

SensatUrban

Urban photogrammetry:

• Aerial and mobile acquisition
• Dense point clouds
• 13 semantic classes

Indoor Datasets

Stanford 2D-3D-S

Indoor spaces:

• RGB-D and LiDAR
• Semantic annotations
• Multiple modalities

ScanNet

RGB-D indoor dataset:

• 1,513 scans
• Semantic voxel labels
• Camera poses

Specialized Datasets

Forest/Vegetation

• TreeLS data
• Forest inventory
• Individual tree detection

Infrastructure

• Bridge inspection
• Tunnel mapping
• Power line detection

Common Tasks

• 3D object detection
• Semantic segmentation
• Instance segmentation
• 3D tracking
• Odometry/SLAM
• HD map creation

Evaluation Metrics

• 3D IoU
• Average Precision (AP)
• Mean Average Precision (mAP)
• Segmentation accuracy
• Tracking metrics (MOTA, MOTP)