ForestMamba: Sparse Mamba with Geometry-guided Queries for 3D Forest Point Cloud Segmentation

Trung Thanh Nguyen, Tuan-Anh Vu, Duc Viet Le, Yasutomo Kawanishi, Takahiro Komamizu, Ichiro Ide, Teja Kattenborn · 2026

Read ForestMamba: Sparse Mamba with Geometry-guided Queries for 3D Forest Point Cloud Segmentation by Trung Thanh Nguyen, Tuan-Anh Vu, Duc Viet Le, Yasutomo Kawanishi, Takahiro Komamizu, Ichiro Ide, Teja Kattenborn (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2606-01549.txt.

Description

Nguyen and co-authors segment 3D forest point clouds with ForestMamba. LiDAR scans turn complex forest structure into usable data. Forest monitoring and biodiversity checks gain from that data. Forest scenes are hard: huge volumes and irregular sampling. Canopy overlap and geographic variation add to the difficulty. Sparse convolutions and transformers do well but scale poorly. Attention cost grows quadratically with scene size. Generic context models also miss forest structural priors.

ForestMamba pairs forest priors with linear-time state-space modeling. A sparse encoder sorts voxels into vertical slabs first. Canopy maxima seed geometry-guided queries for the decoder. A Mamba decoder refines queries with local voxel data. Experiments span seven forest regions.

Catalog record

CREATORTrung Thanh Nguyen, Tuan-Anh Vu, Duc Viet Le, Yasutomo Kawanishi, Takahiro Komamizu, Ichiro Ide, Teja Kattenborn
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE44 KB
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2606.01549

Browse the full library · About the Tower of Babel