Point Cloud Datasets for 3D Perception AI
LiDAR and depth sensor point cloud datasets with 3D bounding box, semantic segmentation, and instance annotation. Built for production 3D perception AI in autonomous systems, robotics, and industrial applications.
20+
Years in AI training data
1,000+
Language locales
1M+
Hours of video annotated
ISO
27001 certified
The Challenge
Beyond Public LiDAR Benchmarks
SemanticKITTI and nuScenes advanced outdoor LiDAR perception research. Production 3D AI for indoor robotics, industrial inspection, and specialized autonomous systems requires point cloud data from your sensor type and environment.
Indoor mobile robots, warehouse automation, and construction site AI operate in environments with dense clutter, limited range, and object types absent from outdoor driving benchmarks. Models trained on automotive LiDAR fail entirely on these applications.
Off-the-shelf point cloud datasets suffer from sensor specificity (each LiDAR model produces different point densities and scan patterns) and environment narrowness (outdoor urban driving dominates public benchmarks).
LXT builds custom point cloud datasets collected with your specific sensor, in your deployment environment, with your object taxonomy. We deliver 3D annotation that trains production-ready 3D perception models for your application.
Why Teams Upgrade
Limitations of Public Point Cloud AI Datasets
Standard benchmarks serve research well. Production deployments need more.
| Dataset | Primary Limitation | Impact |
|---|---|---|
| SemanticKITTI | Outdoor urban driving only; Velodyne HDL-64 sensor; 28 semantic classes fixed to road scenes; no indoor or industrial coverage | Outdoor driving |
| nuScenes-lidarseg | Autonomous vehicle driving; 32-class taxonomy; 32-beam LiDAR only; no non-automotive or indoor applications | AV-only |
| S3DIS | Indoor building scans only; Matterport RGB-D; no dynamic object annotation; limited to static architectural elements | Static indoor |
| ScanNet | RGB-D indoor; limited LiDAR density; no outdoor or industrial coverage; small object annotation quality varies | RGB-D only |
| ShapeNet | Synthetic 3D models only; no real-world scan noise; distribution differs substantially from real sensor point clouds | Synthetic only |
Not sure which specs you need?
Our data specialists help you scope the right dataset for your model architecture.
Configurable Specifications
Specs Built Around Your Model
Public datasets come fixed. Yours is configured for your architecture, environment, and use case.
Sensor Configuration
Hardware Setup
- LiDAR Models: Velodyne, Ouster, Hesai, Livox, and domain-specific solid-state LiDAR
- Depth Sensors: Intel RealSense, Azure Kinect, and structured light depth cameras
- Sensor Fusion: Synchronized LiDAR and camera for sensor fusion model training
3D Annotation Types
Label Coverage
- 3D Bounding Boxes: Axis-aligned and oriented 3D boxes for object detection models
- Semantic Segmentation: Per-point class labels for scene understanding models
- Instance Segmentation: Per-point instance IDs for panoptic 3D segmentation
Environment Coverage
Scene Diversity
- Indoor: Warehouses, factories, buildings, and structured indoor environments
- Outdoor: Roads, construction sites, ports, and specialized outdoor settings
- Conditions: Varying point density, weather effects, and sensor range conditions
Need a custom configuration?
We've built datasets across dozens of domains and use cases. Let's scope yours.
Capturing Complexity
Edge Cases in Point Cloud Annotation
High-accuracy models handle rare attributes that public datasets miss.
Low-Density Distant Objects
Objects at sensor range limits produce sparse point clusters that challenge detection. Annotation guidelines define minimum point count thresholds per distance tier with explicit low-density labeling.
Overlapping and Occluded Objects
Dense scenes produce partially occluded object point clouds. Occlusion level flags and amodal bounding box annotations support occlusion-robust 3D detection training.
Dynamic Objects in Motion
Moving objects in spinning LiDAR scans produce motion-distorted point clouds. Motion distortion flags and corrected annotation versions are provided for motion-sensitive applications.
Reflective and Absorptive Surfaces
Metallic and black surfaces produce LiDAR return artifacts. Domain-specific surface material flags annotate sensor artifact regions to prevent training on corrupted measurements.
Ground Truth Quality
Human-in-the-Loop Annotation
Precise annotation bridges raw data and learnable signal. Expert annotators deliver precision automated tools can't match.
3D Bounding Box Annotation
Expert annotators place oriented 3D bounding boxes with class labels and attribute flags. Annotation tools with point cloud visualization ensure sub-voxel placement accuracy.
Semantic Point Labels
Per-point semantic class assignment for scene segmentation. Quality verified by per-class IoU metrics against held-out review frames.
Instance and Panoptic Labels
Per-point instance IDs alongside semantic labels for panoptic 3D segmentation models supporting both thing and stuff class annotation.
Industry Applications
Point Cloud Datasets for Your Domain
Custom taxonomies and collection protocols for specific deployment contexts.
Autonomous Driving
3D vehicle, pedestrian, and infrastructure detection
Robotics
Indoor navigation, object manipulation, SLAM
Industrial AI
Factory floor mapping, part detection, inspection
Drone AI
Aerial LiDAR, terrain and structure mapping
Construction AI
Site mapping, progress monitoring, safety
Geospatial AI
Urban mapping, forestry, infrastructure survey
Maritime AI
Port and vessel 3D perception
Scientific AI
3D scanning, measurement, research applications
Compliance & Ethics
Secure and Ethical Data Collection
Data collection involving people and sensitive content requires robust security, compliance, and ethical protocols at every stage.
Global Demographic Reach
Collection across 1,000+ locales and diverse demographics to prevent algorithmic bias in your deployed models.
ISO 27001 Certified
Sensitive projects processed in certified secure facilities meeting the highest information security standards.
GDPR & Privacy Compliance
All collection and annotation protocols vetted for consent and privacy. Legally robust for global deployment.
Frequently Asked Questions
Point Cloud Dataset FAQs
Can you collect with our specific LiDAR sensor?+
Yes. We configure collection rigs for your specific sensor model to ensure point density, scan pattern, and noise characteristics match your deployment system.
What 3D annotation formats do you deliver?+
KITTI bin + label, nuScenes JSON, Open3D PCD, and custom formats. 3D bounding box coordinates in sensor and world frames with calibration matrices.
Can you annotate our existing point cloud recordings?+
Yes. We annotate existing LiDAR recordings under a data processing agreement using your object taxonomy and annotation guidelines.
Do you support multi-return and intensity annotations?+
Yes. Multi-return point clouds with intensity values are annotated with return index and intensity metadata preserved in delivery files.
What does a custom point cloud dataset cost?+
Projects range from $20K for focused single-environment collections (1,000-5,000 frames) to $200K+ for large multi-environment, multi-sensor collections with full panoptic annotation.
Can you provide sensor fusion datasets with synchronized camera?+
Yes. Synchronized LiDAR and camera collection with calibration matrices for multi-modal 3D detection and segmentation model training.
How do you handle dynamic scene annotation for moving objects?+
We use multi-frame context annotation for moving objects and provide motion vectors and track IDs for multi-object tracking model training.
Get Started
Scope Your Custom Point Cloud Dataset
Share your sensor type, environment, and 3D annotation requirements. A 3D AI data specialist will provide a detailed proposal within 48 hours.
Please provide us with the details of your inquiry and one of our team members will be in touch.
Join our global team of contributors today
Apply here to be considered for future projects including data collection, annotation and transcription
Start application