Classified out of the box
Road surface, vegetation, structures, vehicles and more — labeled automatically using SAM3, YOLO-class and RF-DETR architecture models.
Semantic segmentation
AI-labeled point clouds. Skip manual annotation.
LiDAR semantic annotation is, in the academic literature's own words, notoriously expensive and time-consuming. LidarFlow runs modern segmentation models on your map so you start with a classified, queryable dataset instead of raw geometry.
Queryable dataset
Filter, export, or feed only the classes you care about into the next step in your pipeline. The geometry stays, the noise filters out.
No per-point annotation budget
Removes the need for expensive manual labeling. Reserve human review for edge cases, not every cubic meter.
Feeds the rest of your stack
Train perception models, auto-extract HD-map features like lanes, poles and signs, or filter dynamic objects out before re-running SLAM.
Classified points, not just geometry
Get a labeled map on the same run that produced the geometry.
Related: LiDAR SLAM, Autoware export.