This repository contains the code release for the CVPR 2023 project
FreeNeRF: Improving Few-shot Neural Rendering with Free Frequency Regularization,
Jiawei Yang, Marco Pavone, and Yue Wang
Computer Vision and Pattern Recognition (CVPR), 2023
The code is based on RegNeRF implementation.
Novel view synthesis with sparse inputs is a challenging problem for neural radiance fields (NeRF). Recent efforts alleviate this challenge by introducing external supervision, such as pre-trained models and extra depth signals, and by non-trivial patch-based rendering. In this paper, we present Frequency regularized NeRF (FreeNeRF), a surprisingly simple baseline that outperforms previous methods with minimal modifications to the plain NeRF. We analyze the key challenges in few-shot neural rendering and find that frequency plays an important role in NeRF's training. Based on the analysis, we propose two regularization terms. One is to regularize the frequency range of NeRF's inputs, while the other is to penalize the near-camera density fields. Both techniques are ``free lunches'' at no additional computational cost. We demonstrate that even with one line of code change, the original NeRF can achieve similar performance as other complicated methods in the few-shot setting. FreeNeRF achieves state-of-the-art performance across diverse datasets, including Blender, DTU, and LLFF. We hope this simple baseline will motivate a rethinking of the fundamental role of frequency in NeRF's training under the low-data regime and beyond.
TL;DR: We use frequency regularization and occlusion regularization to improve few-shot neural rendering. Both techniques can be implemented with a few lines of code.
We recommend to use Anaconda to set up the environment. First, create a new freenerf environment:
conda create -n freenerf python=3.6.15
Next, activate the environment:
conda activate freenerf
You can then install the dependencies:
pip install -r requirements.txt
Finally, install jaxlib with the correct CUDA version. For example, if you have CUDA 11.0 installed, you can run
pip install --upgrade pip
pip install --upgrade jaxlib==0.1.68+cuda110 -f https://storage.googleapis.com/jax-releases/jax_cuda_releases.html
Please follow RegNeRF's data preparation instructions to prepare the DTU and LLFF datasets.
For training a new model from scratch, you need to first need to define your CUDA devices. For example, when having access to 4 GPUs, you can run
export CUDA_VISIBLE_DEVICES=0,1,2,3
and then you can start the training process by calling
python train.py --gin_configs configs/{CONFIG}
where you replace {CONFIG} with the config you want to use.
We provide the config files for FreeNeRF, RegNeRF, and our direct baseline mipNeRF in the configs folder. In the config files, you might need to adjust the Config.data_dir argument pointing to your dataset location. For the DTU dataset, you might further need to adjust the Config.dtu_mask_path argument.
We provide the wandb interface for logging. You might need to install wandb and change the the wandb settings in inter/configs.py to fit your own wandb account.
Alternatively, we provide some sample scripts for training our models in the sample_scripts folder. For example, you can run
bash sample_scripts/train_eval_dtu.sh freenerf 3 30
where freenerf is the method name, 3 is the number of shots, and 30 is the scan ID. Please refer to the sample_scripts folder for more details.
You can render and evaluate test images by running
python eval.py --gin_configs configs/{CONFIG}
where you replace {CONFIG} with the config you want to use. Similarly, you can render a camera trajectory (which we used for our videos) by running
python render.py --gin_configs configs/{CONFIG}
Our results are obtained on 4 NVIDIA A40 GPUs. The results can be slightly different across different runs due to the randomness in the training process. For example, in the DTU dataset, we can obtain a PSNR as high as 20.21 dB under the 3-shot setting, compared to 19.92 dB in the paper.
Please find the results and models in this google drive. Note that these results are reproduced versions and might not strictly be the same as those on our project page. In addition, we use the reg_end_ratio in the provided config files instead of the explicit reg_end iterations in this code.
- We will release the pre-trained models soon.
If you find our work useful, please cite it as
@InProceedings{Yang2023FreeNeRF,
author = {Jiawei Yang and Marco Pavone and Yue Wang},},
title = {FreeNeRF: Improving Few-shot Neural Rendering with Free Frequency Regularization},
booktitle = {Proc. IEEE Conf. on Computer Vision and Pattern Recognition (CVPR)},
year = {2023},
}