README.md

https://github.com/user-attachments/assets/b0752f15-c941-46ad-8996-ea80316a5482

:bulb: Highlight

• :heart_eyes: :heart_eyes: SplatWeaver is a feed-forward framework that adaptively allocates Gaussian primitives based on local scene complexity. By concentrating primitives in intricate regions while maintaining sparsity in smooth areas, our approach enables a more principled and flexible allocation of Gaussians within the scene, yielding superior rendering quality with fewer primitives.

:label: TODO

• Paper & Demo Video.
• Pretrained Models and Inference Demo.
• Training and Evaluation Demo.
• Full Training Implementation.

Installation

1. Clone SplatWeaver.

git clone https://github.com/yecongwan/SplatWeaver.git
cd SplatWeaver

2. Create the environment.

conda create -y -n splatweaver python=3.10
conda activate splatweaver
pip install torch==2.2.0 torchvision==0.17.0 torchaudio==2.2.0 --index-url https://download.pytorch.org/whl/cu121
pip install -r requirements.txt

Quick Start

# Inference with the example in the folder “examples/garden”

python demo.py

Training

# dataset preprocessing

Please prepare data in the datasets folder. You can add any in-the-wild multi-view data with the general dataset class.


# start training:
python src/main +experiment=multi-dataset trainer.num_nodes=1

Evaluation

# Novel View Synthesis
python src/eval_nvs.py --data_dir ... --ckpt_path ...

Citation

If you find the code helpful in your research or work, please cite the following paper:

@article{wan2026splatweaver,
  title={SplatWeaver: Learning to Allocate Gaussian Primitives for Generalizable Novel View Synthesis},
  author={Wan, Yecong and Li, Fan and Shao, Mingwen and Zuo, Wangmeng},
  journal={arXiv preprint arXiv:2605.07287},
  year={2026}
}

Acknowledgement

We thank all authors behind these repositories for their excellent work: VGGT, AnySplat, and gsplat.