OctFormer

Segments large 3D point clouds using an octree to keep attention tractable.

Field
Cross-domain
Method
Deep learning
Data it takes
point-cloud
License
MIT

Transformer attention scales poorly with point count. OctFormer partitions the cloud into an octree so attention operates within local regions, which keeps memory manageable on scans too large for a dense transformer.

Relevant when a scan is big enough that other architectures run out of GPU memory before they run out of accuracy.

Upstream project →

Catalog entry last checked 2026-08-19. All models →

Run OctFormer on your data

The lab runs this on its own compute, in a container, with the inputs and parameters recorded alongside the result. Initial scoping conversations are free.

Contact the Lab