Knowledge_NeRF

This is the official implementation for Knowledge NeRF: Few-shot Novel View Synthesis for Dynamic Articulated Objects. [arXiv]

Install

Please refer to NeRF-pytorch.

Datasets

We propose a new dynamic dataset consisting of synthetic and real-world articulated objects, consisting of 2 spaces: original space and deformed space. If you are interested in our dataset, please reach out to russ000robin@gmail.com.

NeRF Synthetic Dataset and Shiny Blender Dataset are also adopted in our paper.

Run

Our method includes 3 steps:

1. Initialize projection module
2. Train projection module
3. Finetune projection module and original NeRF

Please bash pipeline.sh to run Knowledge NeRF.

Citation

For any questions, please feel free to reach out to start an issue in this repo.

@article{cai2024knowledge,
  title={Knowledge NeRF: Few-shot Novel View Synthesis for Dynamic Articulated Objects},
  author={Cai, Wenxiao and Lei{\i}nst, Xinyue and He, Xinyu and Chen, Junming Leo and Wang, Yangang},
  journal={arXiv preprint arXiv:2404.00674},
  year={2024}
}