Geometry-aware Texture Transfer for Gaussian Splatting

This repository contains the official implementation of the paper "Geometry-aware Texture Transfer for Gaussian Splatting", introducing a novel approach for texture transfer in 3D scenes represented by Gaussian Splatting.

Update🔥

• [2025/11/08] GT²-GS has been accepted by AAAI 2026! Code is coming soon!
• [2025/05/22] We release the paper on arXiv.

Installation

Requirements

• NVIDIA GPU with CUDA 11.8
• Python 3.10
• PyTorch 2.3.0

Conda

# Clone repository with submodules
git clone https://github.com/vpx-ecnu/GT2-GS --recursive
cd GT2-GS

# Install Python dependencies
conda env create -f environment.yaml
conda activate GT2-GS
pip install gs/submodules/diff-gaussian-rasterization
pip install gs/submodules/simple-knn

Quick Start

Dataset and Checkpoint

• For scene dataset, you can find LLFF dataset in NeRF and T&T dataset in 3DGS.
• For style dataset, you can find it in here.
• For texture dataset, you can find it in here. #TODO
• For optimal stylization results, ensure that the original scene is trained using 0th-order spherical harmonics (SH) coefficients. Higher-order SH coefficients may introduce artifacts or inconsistencies during the style transfer process. Using 0th-order SH coefficients ensures smoother and more coherent stylization.

Texture Transfer

python style_main.py --config configs/llff_texture.yaml --stylized_model_path ./output/texture/llff/fern
python scripts/render_llff_video.py --config ./output/texture/llff/fern/config.yaml

Style Transfer

python style_main.py --config configs/llff_style.yaml --stylized_model_path ./output/llff/fern
python scripts/render_llff_video.py --config ./output/llff/fern/config.yaml

Please check python style_main.py --help or files under configs/ for help.

Contact

If you have any questions or suggestions, feel free to open an issue on GitHub. You can also contact Garv1tum and lzlcs directly.

Citation

If you find this project useful, please give a star⭐ to this repo and cite our paper:

@inproceedings{liu2026gt2,
  title={GT2-GS: Geometry-aware Texture Transfer for Gaussian Splatting},
  author={Liu, Wenjie and Liu, Zhongliang and Shu, Junwei and Wang, Changbo and Li, Yang},
  booktitle={Proceedings of the AAAI Conference on Artificial Intelligence},
  volume={40},
  number={9},
  pages={7296--7304},
  year={2026}
}