This repository contains a pytorch implementation for the paper: FourieRF: Few-Shot NeRFs via Progressive Fourier Frequency Control. We present a simple yet efficient approach to tackle the few-shot NeRF problem. This repository is built on top of the code base introduced by the paper TensoRF: Tensorial Radiance Fields.
output.mp4
conda create -n FourieRF python=3.10 -y
conda activate FourieRF
pip install torch torchvision
pip install tqdm scikit-image opencv-python configargparse lpips imageio-ffmpeg kornia lpips tensorboard scikit-learn plyfile matplotlib
Correctly set the data path on the configs/fourier_blender.txt file with the datadir argument.
The training script is in train.py, to train a FourieRF on the orchids scene of the forward-facing dataset:
python train.py --config configs/fourier_blender.txt --number_of_views 4
If you find our code or paper helps, please consider citing:
@misc{gomez2025fourierffewshotnerfsprogressive,
title={FourieRF: Few-Shot NeRFs via Progressive Fourier Frequency Control},
author={Diego Gomez and Bingchen Gong and Maks Ovsjanikov},
year={2025},
eprint={2502.01405},
archivePrefix={arXiv},
primaryClass={cs.CV},
url={https://arxiv.org/abs/2502.01405},
}
