# manycore-research/SpatialLM

[NeurIPS 2025] SpatialLM: Training Large Language Models for Structured Indoor Modeling

Repository: https://github.com/manycore-research/SpatialLM
Canonical: https://ross.abutalabs.com/products/spatiallm
Homepage: https://manycore-research.github.io/SpatialLM
Language: Python
License: NOASSERTION
License Family: other
Topics: scene-understanding, spatial-intelligence, mllm, point-clouds
Last push: 2026-06-26T18:14:20+00:00

## Health v2 (maintenance only)
Score: 62/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 89, release rhythm 40, longevity 38
- inputs: {"age_days": 538, "days_push": 68, "days_rel": 449, "gap_med": 0, "n_releases_24m": 2}
- flags: no_license
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 4719, forks 401 (observed 2026-08-28T04:08:57.480559+00:00)

## What it is
SpatialLM is a 3D large language model that processes point cloud data (from monocular video, RGBD images, or LiDAR) and generates structured 3D scene understanding outputs such as walls, doors, windows, and oriented object bounding boxes with semantic categories. It includes pretrained models, a training dataset, and finetuning instructions.

## Use cases
- reconstruct indoor 3D layout from a monocular RGB video
- detect walls doors and objects in a 3D point cloud
- convert point clouds into structured scene representations
- spatial reasoning for embodied robotics and navigation
- finetune a 3D scene understanding LLM on custom data

## When to choose
- you need structured indoor scene understanding from point clouds or video
- you want semantic 3D layouts without specialized scanning equipment
- you're researching spatial intelligence or multimodal LLMs

## When to avoid
- you need outdoor or large-scale scene reconstruction
- you need a lightweight real-time pipeline without GPU inference
- you need a permissively licensed model for commercial use without checking the custom license

## Facets
- artifact type: library
- maturity: active
- function: machine-learning, llm-inference, computer-vision, nlp
- domain: computer-vision, large-language-models, artificial-intelligence, robotics
- platform: python
- tags: point-cloud, 3d-scene-understanding, spatial-intelligence, indoor-scene-reconstruction, multimodal-llm, scene-layout, gpu, linux

## Member repositories
- manycore-research/SpatialLM (main) score 62

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:08:57.480559+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-29T18:19:06.357223+00:00, confidence not recorded.
  - readme: https://github.com/manycore-research/SpatialLM (fetched 2026-08-28T04:08:57.480559+00:00, sha d37c511bda1a)
  - homepage: https://manycore-research.github.io/SpatialLM (fetched 2026-08-29T09:02:53.940494+00:00, sha 43bf5f60285e)
- Data as of 2026-08-30T08:39:29.467469+00:00.