SLNR

A Super Lightweight Neural Representation for Large-scale 3D Mapping
Chenhui Shi, Fulin Tang, Hao Wei, Yihong Wu

Abstract: We propose a new and ultra-lightweight neural representation with outstanding performance for large-scale 3D mapping. The representation defines a global signed distance function (SDF) in near-surface space based on a set of band-limited local SDFs anchored at support points sampled from point clouds. These SDFs are parameterized only by a tiny multi-layer perceptron (MLP) with no latent features, and the state of each SDF is modulated by three learnable geometric properties: position, rotation, and scaling, which make the representation adapt to complex geometries. Then, we develop a novel parallel algorithm tailored for this unordered representation to efficiently detect local SDFs where each sampled point is located, allowing for real-time updates of local SDF states during training. Additionally, a prune-and-expand strategy is introduced to enhance adaptability further. The synergy of our low-parameter model and its adaptive capabilities results in an extremely compact representation with excellent expressiveness.

Introduce

This repository provides four core components:

• A local SDF representation without relying on latent features;
• A parallel local SDF detection algorithm for real-time optimization;
• A prune-expand strategy to enhance the adaptability;
• Parallel point sampling along rays in a spatial hash grid.

Install

The code has been tested on an RTX 4090 GPU with CUDA-11.8 equipped. First, install the Python environment:

conda create --name slnr python=3.8
pip install torch==2.0.0 torchvision==0.15.1 torchaudio==2.0.1 --index-url https://download.pytorch.org/whl/cu118
pip install open3d pypose opencv-python scikit-image

Then, build the custom modules:

# Related to the spatial hash grid
cd ./modules/sparse_hash && mkdir build && cd build
cmake -DCMAKE_PREFIX_PATH=your_env/lib/python3.8/site-packages/torch  ..
make -j$(nproc)
# Related to the local sdf detection and optimization
cd ./modules/gaussian_search && mkdir build && cd build
cmake -DCMAKE_PREFIX_PATH=your_env/lib/python3.8/site-packages/torch  ..
make -j$(nproc)

Note that in ./Thirdparty/skimage , we provide a modified version of scikit-image. Specifically, changes have been made at lines 161 and 187 of measure/_marching_cubes_lewiner.py to ignore regions without surfaces during the marching cubes process.

Run

First, download the Oxford Spires Example Data. Please refer to the official website for more detialed information and observe the license. Then, put the data to ./data folder. Finally, execute the following command:

python run.py --conf=configs/example.yaml

The real-time optimization for basis SDF and geometric parameters is presented as below. Each support point is colored by the z-axis orientation of the local coordinate system of the defined local SDF.

Some results are shown as below:

Acknowledgement

This work was inspired by the B-spline basis function used in PoissonRecon, the neural point representation in PIN-SLAM and the splatting algorithm in 3DGS. We hope that this work will inspire interested parties to develop more lightweight and efficient mapping methods for robotic spatial intelligence.

Citation

@article{shi2026slnmapping,
  title={SLNMapping: Super Lightweight Neural Mapping in Large-Scale Scenes},
  author={Shi, Chenhui and Tang, Fulin and Wei, Hao and Wu, Yihong},
  journal={International Journal of Computer Vision},
  volume={134},
  number={2},
  pages={68},
  year={2026},
  publisher={Springer}
}

@inproceedings{shi20253d,
  title={3D-SLNR: A Super Lightweight Neural Representation for Large-scale 3D Mapping},
  author={Shi, Chenhui and Tang, Fulin and An, Ning and Wu, Yihong},
  booktitle={Proceedings of the Computer Vision and Pattern Recognition Conference},
  pages={27233--27242},
  year={2025}
}

If using the component of parallel point sampling along rays in a spatial hash grid, please cite:

@article{shi2023accurate,
  title={Accurate implicit neural mapping with more compact representation in large-scale scenes using ranging data},
  author={Shi, Chenhui and Tang, Fulin and Wu, Yihong and Jin, Xin and Ma, Gang},
  journal={IEEE Robotics and Automation Letters},
  volume={8},
  number={10},
  pages={6683--6690},
  year={2023},
  publisher={IEEE}
}