The State of Automatic Point Cloud Classification in 2026
How deep learning has finally made ground, vegetation and building classification a solved problem — and what surveyors should do next.
How deep learning has finally made ground, vegetation and building classification a solved problem — and what surveyors should do next.

Five years ago, classifying a mobile mapping point cloud was a weeks-long, half-manual chore. Today, VisionLidar's classifier processes a full corridor in an afternoon — and gets it right the first time.
The shift came from two changes: transformer-based segmentation networks that see local geometry the way a surveyor does, and training datasets large enough to cover every scanner brand on the market.
In this post we break down what to expect from modern classifiers, where they still struggle (dense vegetation over water, we're looking at you), and how to build a QA workflow that trusts automation without turning off your brain.
Book a 30-minute demo with a senior product specialist, tailored to your industry.