Neural Texture Splatting: Expressive 3D Gaussian Splatting for View Synthesis, Geometry, and Dynamic Reconstruction

SIGGRAPH Asia 2025 (Conference Track)

Yiming Wang, Shaofei Wang, Marko Mihajlovic, Siyu Tang

ETH Zürich

Neural Texture Splatting (NTS) extends 3D Gaussian Splatting by introducing a local neural RGBA field per primitive. This codebase is built on Gaussian Opacity Fields, showcasing NTS’s ability to improve novel-view synthesis and surface reconstruction.

Setup

1. Environment

conda create -y -n nts python=3.8
conda activate nts

# Install PyTorch (Adjust for your CUDA version if necessary)
pip install torch==2.0.1 torchvision==0.15.2 --index-url [https://download.pytorch.org/whl/cu118](https://download.pytorch.org/whl/cu118)
conda install -y -c "nvidia/label/cuda-11.8.0" cuda-toolkit

# Install dependencies
pip install -r requirements.txt
pip install submodules/diff-gaussian-rasterization_3dtex
pip install submodules/simple-knn/

Optional: Build tetra-triangulation (required for GOF-style mesh extraction). If you encounter issues, please check the original GOF repository.

cd submodules/tetra-triangulation
conda install cmake
conda install conda-forge::gmp
conda install conda-forge::cgal

cmake .
# (Optional) Specify your CUDA path:
# export CPATH=/usr/local/cuda-11.3/targets/x86_64-linux/include:$CPATH

make
pip install -e .

2. Data Preparation

• NeRF-Synthetic: Download from the NeRF Google Drive.
• DTU:
  • Images: Download the preprocessed data from the 2DGS website.
  • GT Point Clouds: Download from the official DTU website (required for geometry evaluation).

Training and Evaluation

Use these scripts to reproduce the experiments reported in the paper.

# NeRF
python scripts/run_nerf_synthetic.py
python scripts/show_nerf_synthetic.py

# DTU
python scripts/run_dtu.py
python scfipts/show_dtu.py

TODO

• 3DGS-MCMC + NTS for improved unbounded scene performance.
• SplatFields + NTS for sparse-view and dynamic reconstruction.

Citation

@misc{wang2025neuraltexturesplattingexpressive,
      title={Neural Texture Splatting: Expressive 3D Gaussian Splatting for View Synthesis, Geometry, and Dynamic Reconstruction}, 
      author={Yiming Wang and Shaofei Wang and Marko Mihajlovic and Siyu Tang},
      year={2025},
      eprint={2511.18873},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/2511.18873}, 
}