SVD for 3DGS-Enhancer

Stable Video Diffusion Training Code for 3DGS-Enhancer

Training the SVD for 3DGS-Enhancer

To train the model, run the following command:

sbatch train_single_nodes.sh

Inference with the SVD for 3DGS-Enhancer

To perform inference, use the following command:

sbatch multi_nodes_inference.sh

Note

Before using SVD to generate results, ensure you use split_scene_to_txt.py to generate a folder containing text files that are written by scene names.

Acknowledgements

This work builds upon the following codebases:

• SVD_Xtends
• Diffusers

We thank the authors of these projects for their excellent contributions!

Citation

If you use 3DGS-Enhancer in your research, please cite the following paper:

@article{liu20243dgs,
  title={3DGS-Enhancer: Enhancing unbounded 3D Gaussian splatting with view-consistent 2D diffusion priors},
  author={Liu, Xi and Zhou, Chaoyi and Huang, Siyu},
  journal={arXiv preprint arXiv:2410.16266},
  year={2024}
}