Skip to content

Arxiv 2023 : boosting performance of few-shot NeRF models with self-supervised training

License

Notifications You must be signed in to change notification settings

crepejung00/SE-NeRF

 
 

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

14 Commits
 
 
 
 
 
 
 
 

Repository files navigation

SE-NeRF : Self-Evolving Neural Radiance Fields


This is our official implementation of Self-Evolving Neural Radiance Fields!

by Jaewoo Jung*, Jisang Han*, Jiwon Kang*, Seongchan Kim, Min-seop Kwak, Seungryong Kim

Updates

⚠ Please notice that the code is currently being refactored, so minor bugs or mismatches with the metrics in the original paper might exist. Please feel free to raise any issues and we will fix it as soon as possible!

2024/02 : We released our codes for SE-NeRF(K-Planes) on the NeRF Synthetic & NeRF Synthetic Extreme dataset!

Introduction

We introduce a self-training framework applicable to existing Neural Radiance Fields!

We show that the self-training strategy successfully mitigates the overfitting problem when training NeRFs with only sparse inputs.

For further details and visualization results, please check out our paper and our project page.

Installation

Please follow installation.

Datasets

Training and evaluation data can be downloaded from the respective websites.

NeRF Synthetic Dataset(blender) : Data.

Training

Currently only K-Planes is the supported backbone model!

To train the model with our framework, all you need to do is:

python train.py --backbone kplanes

Acknowledgement

As we show that our framework succesfully applies self-training to existing NeRF models, we utilized the public codes for the baseline networks in this repository.

We would like to acknowledge the contributions of K-Planes for open-sourcing the official codes for K-Planes!

Citation

@misc{jung2023selfevolving,
      title={Self-Evolving Neural Radiance Fields}, 
      author={Jaewoo Jung and Jisang Han and Jiwon Kang and Seongchan Kim and Min-Seop Kwak and Seungryong Kim},
      year={2023},
      eprint={2312.01003},
      archivePrefix={arXiv},
      primaryClass={cs.CV}
}

About

Arxiv 2023 : boosting performance of few-shot NeRF models with self-supervised training

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published