Segment Anything
Segment Anything Model (SAM): A new AI model that can "cut out" any object, in any image, with a single click.
Teach AI what it sees.
Label 3D point clouds, images, and video in one tool, fused across sensors, accelerated by AI, and verified by people. The verified data perception models actually learn from.
From one click to a finished label.
Propagate Labels
Propagate Labels in Semantic Segmentation with just a few clicks.
Superpixel
Pixel-accurate ML-assisted segmentation, segment an object with just a few clicks.
Bounding box & segmentation
Pixel-wise object labeling by drawing a bounding box.
Frames Classification
Pre-label up to 80 common classes automatically by clicking on auto-label. Improve your productivity by 7x.
Label one frame. Track the whole sequence.
Visual Object tracking
Accurate object detection and tracking across frames.
Frames Classification
Pre-label up to 80 common classes automatically by clicking on auto-label. Improve your productivity by 7x.
Seamless labeling navigation
Drag to delete and add labels across the sequence, and delete any label at any frame with a click.
Your ontology. Your formats.
ASAM OpenLabelLeRobotROS 2MCAPnuScenesHDF5ParquetCOCORosbagZarrKITTIORCPascalWebDataset
Fused points, edited from every view.
Fused Cloud
Once the Fused cloud is selected, the accumulated points in the lidar will be shown in the current frame. The points will be shown only for the frames which are preloaded
Label-view
View selected labels across all frames - side view, front view and back view. Can also be directly edited on the go.
Frames Classification
Pre-label up to 80 common classes automatically by clicking on auto-label. Improve your productivity by 7x.
Supported Annotation Types
Comprehensive labeling capabilities for every perception use case
2D Annotation
- Bounding boxes & polygons
- Semantic/instance segmentation
- Keypoints & landmarks
3D & LiDAR Annotation
- 3D bounding boxes
- Point cloud segmentation
- 3D Polylines & Polygons
- Sequence labeling & multi-frame tracking
Multi-Sensor & Fusion
- Cross-sensor association
- LiDAR, camera, radar & IMU fusion
- Robust perception datasets
Scenario & Behavior Labeling
- Activities & interactions
- Intent & scenario tagging
- Advanced ADAS & robotics models
Trusted by teams building Physical AI.
Deepen has demonstrated true expertise in working with our LiDAR data. They have provided valuable guidance over the taxonomy, customizing their tools and workflow to our evolving data collection platforms, and generating high-quality ground truth.
Pierre Olivier
CTO
LeddarTech
Formats, accuracy, and who does the labeling.
What data can Annotate label?+
3D point clouds, images and video in one tool: 3D bounding boxes, semantic and instance segmentation, and polylines on LiDAR and radar, plus pixel-accurate detection, keypoints and object tracking on image and video.
How accurate is the labeling?+
Every dataset passes a three-stage review — annotator pass, peer review, and QA lead sign-off — backed by automated validation rules and inter-annotator agreement scoring, and exits at 98.6% first-pass accuracy.
Who does the labeling?+
Our own managed, in-house ML experts, never crowdsourced, working across more than 200 programs in production.
What export formats are supported?+
Deepen authored ASAM OpenLabel, and Annotate exports to it along with LeRobot, ROS 2, MCAP, nuScenes, HDF5, Parquet, COCO, Rosbag, Zarr, KITTI, ORC, Pascal, WebDataset and custom schemas. No migration, no lock-in.
Can I use my own ontology?+
Yes. Bring your own taxonomy or start from a template, label 2D and 3D in one project, and ship to the format your training loop expects.
Label data you can trust.
See how Deepen Annotate fits your perception pipeline.