[CVPR 2025] SplatFlow: Multi-View Rectified Flow Model for 3D Gaussian Splatting Synthesis

This repository contains the official pytorch implementation of the paper: "SplatFlow: Multi-View Rectified Flow Model for 3D Gaussian Splatting Synthesis".

Installation

Our code is tested with Python 3.10, Pytorch 2.4.0, and CUDA 11.8. To install the required packages, run the following command:

pip3 install torch==2.4.0 torchvision==0.19.0 torchaudio==2.4.0 --index-url https://download.pytorch.org/whl/cu118
pip install -r requirements.txt

Model Checkpoints

We have extended the model training period after submission to enhance its performance. Updated model checkpoints are now available, and benchmark results will be revised following the review process. You can download the model checkpoints with the following command:

bash download.sh

Inference

1. 3DGS Generation

export PYTHONPATH=$(pwd)
huggingface-cli login 
python inference/generate.py +experiments=generation inference.generate.prompt="Your prompt here"

Citation

If you find this repository helpful for your project, please consider citing our work. :)

@article{go2024splatflow,
  title={SplatFlow: Multi-View Rectified Flow Model for 3D Gaussian Splatting Synthesis},
  author={Go, Hyojun and Park, Byeongjun and Jang, Jiho and Kim, Jin-Young and Kwon, Soonwoo and Kim, Changick},
  journal={arXiv preprint arXiv:2411.16443},
  year={2024}
}

Acknolwedgement

We thank director3d

TODO:

• Update project page.
• Add model checkpoints
• Code verification.
• Add more details on the README.md
• Add the training script
• Share text annotations for RealEstate10K, DL3DV-10K, MVImgNet, and ACID datasets. (Used in the next work, may be integrated with https://github.com/gohyojun15/VideoRFSplat)