Detection & localization
Localize objects in images, video, and point clouds with 2D boxes and 3D cuboids.
Defensible ground truth
Ground truth that keeps the scene intact.
Turn synchronized camera, LiDAR, and telemetry into reviewed training data for Physical AI. Label in 2D, 3D, and across sequences without losing the context your models need.
Software, targeted model assistance, and optional managed operations run inside one quality workflow.
Sequence 041
Sensors synchronizedMulti-camera video
00:18:42.320
Active track · 047
LiDAR and point clouds
- Label
- Human review
- Quality control
Quality controlReady for release
MCAP and ROS data
LiDAR and point clouds
Multi-camera video
Telemetry and trajectories
Multimodal by design
One synchronized scene. Every ground-truth decision.
Physical AI models learn from motion and context—not isolated frames. Avala keeps camera, point-cloud, and trajectory data aligned as work moves through labeling and review.
Segmentation
Create image masks and point-cloud segmentation with targeted model assistance where supported.
Tracking & classification
Keep tracked objects and 4D polylines consistent across a sequence, then route exceptions into review.
Workflow control
Put quality gates inside the pipeline.
Compose visual, conditional workflows with label schemas, task-specific training, human review, and per-station cost estimates. Add classification consensus or selected spatial consensus analytics where the task supports it.
Define schemas and branching logic
Assist selected box and mask workflows
Route work through human review
Gate the dataset handoff on QC
Release evidence
Sequence 041 / release 014
- Annotator
- operator-27
- Review state
- Accepted
- QC history
- 06 → 07
Traceable quality
Keep the evidence attached to the label.
Track annotator and reviewer attribution, accept or reject states, and 3D track QC history. Native-enabled workflows can also preserve version and producer provenance as annotations change.
Training outputs for Physical AI
The geometry Physical AI teams actually ship.
Explore supported detection, segmentation, geometry, and sequence-tracking workflows—without a general-purpose annotation catalog getting in the way.
Detection & localization
Locate objects in image frames and 3D scenes.
Segmentation
Create class or object regions in camera and point-cloud data.
Scene geometry
Trace task-defined boundaries and sequence geometry.
Tracking & classification
Preserve identity and attributes across supported sequences.
Bring one representative scene
Prove the ground-truth workflow before you scale it.
We’ll map one sensor sample, acceptance contract, review path, and dataset handoff with your team.