S3R-GS: Streamlining the Pipeline for Large-Scale Street Scene Reconstruction

Accelerate reconstruction pipeline for large-scale dynamic street scene.

🖥️ GitHub    |    📑 Paper   

Requirements

Codes was tested with the following dependencies

• Python 3.10
• CUDA 11.8
• PyTorch 2.0.1

Installation

1. Clone the repository

git clone https://github.com/Tom-zgt/S3R-GS.git

2. Create a new conda environment as map4d

conda create --name map4d -y python=3.10
conda activate s3rgs
pip install --upgrade pip
pip install torch==2.0.1+cu118 torchvision==0.15.2+cu118 --extra-index-url https://download.pytorch.org/whl/cu118

conda install -c "nvidia/label/cuda-11.8.0" cuda-toolkit
pip install ninja git+https://github.com/hturki/tiny-cuda-nn.git@ht/res-grid#subdirectory=bindings/torch
pip install nerfstudio==1.0.3
python setup.py develop

Pepare Data

Use our preprocessing scripts to prepare the datasets:

mkdir data
#put dataset in the folder ./data/

• Pepare dataset Argoverse 2 as map4d
• Download the preprocessed KITTI scenes here

Training

Use the generated metadata files to train the model on a specific dataset:

ns-train 4dgf-kitti street --data data/KITTI/tracking/training/metadata_[0001/0002/0006]_cover.pkl --train-split-fraction [0.75/0.5/0.25]

ns-train 4dgf-kitti-800 [(optional) --machine.num-devices 4] street --data data/KITTI/tracking/training/metadata_[0009/0020].pkl --train-split-fraction 0.75

ns-train 4dgf-av2-big --machine.num-devices 8 --pipeline.model.max-num-gaussians 8000000 --pipeline.model.object-grid-log2-hashmap-size 17 street --data data/Argoverse2/metadata_PIT_6180_1620_6310_1780.pkl --voxel-size 0.15
ns-train 4dgf-av2-big --machine.num-devices 8 --pipeline.model.max-num-gaussians 8000000 --pipeline.model.object-grid-log2-hashmap-size 17 street --data data/Argoverse2/metadata_PIT_1100_-50_1220_150.pkl --voxel-size 0.15

Evaluation

We provide the trained checkpoints here

ns-eval --load-config <trained_model_config>

Citation

Please consider citing our work with the following references

@article{zheng2025s3r,
  title={S3R-GS: Streamlining the Pipeline for Large-Scale Street Scene Reconstruction},
  author={Zheng, Guangting and Deng, Jiajun and Chu, Xiaomeng and Yuan, Yu and Li, Houqiang and Zhang, Yanyong},
  journal={arXiv preprint arXiv:2503.08217},
  year={2025}
}

Acknowledgements

This project builds on the great work map4d.