Skip to content

SpatialLM

Turn 3D point clouds into structured scene output

Visit official site

What SpatialLM is

The model is described as a large language model designed for 3D point cloud data, producing structured 3D scene understanding output with semantic categories for building elements and objects. The model handles point clouds from monocular video, RGBD images and LiDAR sensors without dedicated equipment.

What you can do with it

  • Process point clouds from monocular video, RGBD images and LiDAR sources
  • Output building elements such as walls, doors and windows
  • Produce object bounding boxes with semantic categories
  • Visualise point clouds and predicted 3D layouts with Rerun

Application scenarios

  • Analyse a building point cloud to identify doors, windows and wall structures
  • Support real-time environment understanding in robot navigation
  • Build educational software for learning 3D modelling and spatial vision

Who it is for

  • Researchers in 3D spatial understanding
  • Developers building intelligent robots and automation systems
  • Educators teaching 3D modelling and spatial analysis

Pros & cons

✓ What we like

  • Several point cloud sources work without dedicated equipment
  • A simple Python environment setup is documented

! What to watch out for

  • The documented install requires CUDA and other dependencies
  • Scene output follows the semantic categories the model produces

FAQ

Which inputs are supported?

The material lists point clouds from monocular video sequences, RGBD images and LiDAR sensors.

What does the output include?

Building elements such as walls, doors and windows, plus object bounding boxes with semantic categories.

Last reviewed: 2026-11-15

More LLM API platform tools

View all →

How we review