What could be automatically recognized in point clouds? That’s the question we hear in almost every meeting.

The short answer is: almost everything you can recognize with your own eyes.

Walls, pipes, ducts, doors, switches, furniture - if a human can identify it, our system can too.

This is the most important part of our video-to-BIM pipeline: semantic understanding of the 3D data.

We use a mix of fixed-class models and open vocabulary detection. The fixed part gives rock-solid reliability on standard elements. The open vocabulary part is for adapting to new environments, handling edge cases, and integrating new object types without retraining from scratch.

For our clients that means: • Faster onboarding of new object classes • More flexibility across building types and conditions • Less manual labeling or modeling

The result? Raw point clouds become structured, labeled, BIM-ready data without the overhead of manual modeling or template-based workflows.

If you’ve ever stared at a dense 3D scan wondering “what’s what?” - this is the layer that changes that.