SpatialLM1.1-Qwen-0.5B

✨ News

• [Sept, 2025] SpatialLM-Dataset is now available on Hugging Face.
• [Sept, 2025] SpatialLM accepted at NeurIPS 2025.
• [Jun, 2025] Check out our new models: SpatialLM1.1-Llama-1B and SpatialLM1.1-Qwen-0.5B, now available on Hugging Face. SpatialLM1.1 doubles the point cloud resolution, incorporates a more powerful point cloud encoder Sonata and supports detection with user-specified categories.
• [Jun, 2025] SpatialLM Technical Report is now on arXiv.
• [Mar, 2025] We're excited to release the SpatialLM-Llama-1B and SpatialLM-Qwen-0.5B on Hugging Face.
• [Mar, 2025] Initial release of SpatialLM!

Introduction

SpatialLM is a 3D large language model designed to process 3D point cloud data and generate structured 3D scene understanding outputs. These outputs include architectural elements like walls, doors, windows, and oriented object bounding boxes with their semantic categories. Unlike previous methods that require specialized equipment for data collection, SpatialLM can handle point clouds from diverse sources such as monocular video sequences, RGBD images, and LiDAR sensors. This multimodal architecture effectively bridges the gap between unstructured 3D geometric data and structured 3D representations, offering high-level semantic understanding. It enhances spatial reasoning capabilities for applications in embodied robotics, autonomous navigation, and other complex 3D scene analysis tasks.

SpatialLM Models

Usage

Installation

Tested with the following environment:

• Python 3.11
• Pytorch 2.4.1
• CUDA Version 12.4

# clone the repository
git clone https://github.com/manycore-research/SpatialLM.git
cd SpatialLM

# create a conda environment with cuda 12.4
conda create -n spatiallm python=3.11
conda activate spatiallm
conda install -y nvidia/label/cuda-12.4.0::cuda-toolkit conda-forge::sparsehash

# Install dependencies with poetry
pip install poetry && poetry config virtualenvs.create false --local
poetry install
# SpatialLM1.0 dependency
poe install-torchsparse # Building wheel for torchsparse will take a while
# SpatialLM1.1 dependency
poe install-sonata # Building wheel for flash-attn will take a while

Inference

In the current version of SpatialLM, input point clouds are considered axis-aligned where the z-axis is the up axis. This orientation is crucial for maintaining consistency in spatial understanding and scene interpretation across different datasets and applications. Example preprocessed point clouds, reconstructed from RGB videos using MASt3R-SLAM, are available in SpatialLM-Testset.

Download an example point cloud:

huggingface-cli download manycore-research/SpatialLM-Testset pcd/scene0000_00.ply --repo-type dataset --local-dir .

Run inference:

python inference.py --point_cloud pcd/scene0000_00.ply --output scene0000_00.txt --model_path manycore-research/SpatialLM1.1-Qwen-0.5B

Detection with user-specified categories

SpatialLM1.1 supports object detection conditioned on user-specified categories by leveraging the flexibility of LLMs.

SpatialLM1.1 offers three variants of structured indoor modeling tasks:

• Structured Reconstruction: Detect walls, doors, windows, boxes.
• Layout Estimation: Detect walls, doors, windows.
• 3D Object Detection: Detect boxes.

For tasks that include object box estimation, you can specify a subset of the 59 furniture categories, and the model will only predict objects within those specified categories. For example:

python inference.py --point_cloud pcd/scene0000_00.ply --output scene0000_00.txt --model_path manycore-research/SpatialLM1.1-Qwen-0.5B --detect_type object --category bed nightstand

Visualization

Use rerun to visualize the point cloud and the predicted structured 3D layout output:

# Convert the predicted layout to Rerun format
python visualize.py --point_cloud pcd/scene0000_00.ply --layout scene0000_00.txt --save scene0000_00.rrd

# Visualize the point cloud and the predicted layout
rerun scene0000_00.rrd

Evaluation

To evaluate the performance of SpatialLM, we provide eval.py script that reports the benchmark results on the SpatialLM-Testset in the table below in section Benchmark Results.

Download the testset:

huggingface-cli download manycore-research/SpatialLM-Testset --repo-type dataset --local-dir SpatialLM-Testset

Run evaluation:

# Run inference on the PLY point clouds in folder SpatialLM-Testset/pcd with SpatialLM1.1-Qwen-0.5B model
python inference.py --point_cloud SpatialLM-Testset/pcd --output SpatialLM-Testset/pred --model_path manycore-research/SpatialLM1.1-Qwen-0.5B

# Evaluate the predicted layouts
python eval.py --metadata SpatialLM-Testset/test.csv --gt_dir SpatialLM-Testset/layout --pred_dir SpatialLM-Testset/pred --label_mapping SpatialLM-Testset/benchmark_categories.tsv

Example using a custom video

We provide an example of how to use our model to estimate scene layout starting from a RGB video with the newly released SLAM3R in EXAMPLE.md. These steps work for MASt3R-SLAM, and other reconstruction methods as well.

SpatialLM Dataset

The SpatialLM dataset is a large-scale, high-quality synthetic dataset designed by professional 3D designers and used for real-world production. It contains point clouds from 12,328 diverse indoor scenes comprising 54,778 rooms, each paired with rich ground-truth 3D annotations. SpatialLM dataset provides an additional valuable resource for advancing research in indoor scene understanding, 3D perception, and related applications.

For access to photorealistic RGB/Depth/Normal/Semantic/Instance panoramic renderings and camera trajectories used to generate the SpatialLM point clouds, please refer to the SpatialGen project for more details.

SpatialLM Testset

We provide a test set of 107 preprocessed point clouds, reconstructed from RGB videos using MASt3R-SLAM. SpatialLM-Testset is quite challenging compared to prior clean RGBD scans datasets due to the noises and occlusions in the point clouds reconstructed from monocular RGB videos.

Benchmark Results

Layout Estimation

Layout estimation focuses on predicting architectural elements, i.e., walls, doors, and windows, within an indoor scene. We evaluated this task on the Structured3D dataset. For RoomFormer, we directly downloaded the model checkpoint. SceneScript and SpatialLM were first trained on our dataset, and further fine-tuned on Structured3D.

3D Object Detection

We evaluate 3D object detection on ScanNet with annotations of 18 object categories. For V-DETR, we directly download the model checkpoint. SceneScript and SpatialLM were first trained on our dataset, and further fine-tuned on ScanNet.

Zero-shot Detection on Videos

Zero-shot detection results on the challenging SpatialLM-Testset are reported in the following table:

Result Visualizations

Layout Estimation	Object Detection	Zero-shot Reconstruction
Structured3D Results	ScanNet Results	Zero-shot Results

License

SpatialLM-Llama-1B is derived from Llama3.2-1B-Instruct, which is licensed under the Llama3.2 license. SpatialLM-Qwen-0.5B is derived from the Qwen-2.5 series, originally licensed under the Apache 2.0 License.

SpatialLM1.0 are built upon the SceneScript point cloud encoder, licensed under the CC-BY-NC-4.0 License. TorchSparse, utilized in this project, is licensed under the MIT License.

SpatialLM1.1 are built upon Sonata point cloud encoder, model weight is licensed under the CC-BY-NC-4.0 License. Code built on Pointcept is licensed under the Apache 2.0 License.

Citation

If you find this work useful, please consider citing:

@inproceedings{SpatialLM,
  title     = {SpatialLM: Training Large Language Models for Structured Indoor Modeling},
  author    = {Mao, Yongsen and Zhong, Junhao and Fang, Chuan and Zheng, Jia and Tang, Rui and Zhu, Hao and Tan, Ping and Zhou, Zihan},
  booktitle = {Advances in Neural Information Processing Systems},
  year      = {2025}
}

Acknowledgements

We would like to thank the following projects that made this work possible:

Llama3.2 | Qwen2.5 | Transformers | SceneScript | TorchSparse | Sonata | Pointcept