😈BAD-RFs: Bundle-adjusted Radience Fields from degraded images with continuous-time motion models

This repo contains an accelerated reimplementation of our CVPR paper BAD-NeRF: Bundle Adjusted Deblur Neural Radiance Fields, based on the nerfstudio framework.

In the future, we will continue to explore bundle-adjusted radience fields, add more accelerated implementations to this repo, such as a reimplementation of our ICLR paper USB-NeRF: Unrolling Shutter Bundle Adjusted Neural Radiance Fields.

Demo

Deblurring & novel-view synthesis results on Deblur-NeRF's real-world motion-blurred data:

ball.mp4

decoration.mp4

girl.mp4

heron.mp4

Left: BAD-NeRFacto deblured novel-view renderings;

Right: Input images.

Quickstart

1. Installation

You may check out the original nerfstudio repo for prerequisites and dependencies. Currently, our codebase is build on top of the latest version of nerfstudio (v1.0.2), so if you have an older version of nerfstudio installed, please git clone the main branch and install the latest version.

TL;DR: You can install nerfstudio with:

# (Optional) create a fresh conda env
conda create --name nerfstudio -y python=3.10
conda activate nerfstudio

# install dependencies
pip install --upgrade pip setuptools
pip install torch==2.1.2+cu118 torchvision==0.16.2+cu118 --extra-index-url https://download.pytorch.org/whl/cu118

conda install -c "nvidia/label/cuda-11.8.0" cuda-toolkit
pip install ninja git+https://github.com/NVlabs/tiny-cuda-nn/#subdirectory=bindings/torch

# install nerfstudio!
git clone https://github.com/nerfstudio-project/nerfstudio
cd nerfstudio
pip install -e .

Besides, we use pypose to implement the pose interpolation. You can install it with:

pip install pypose

Then you can clone and install this repo as a python package with:

git clone https://github.com/WU-CVGL/Bad-RFs
cd Bad-RFs
pip install -e .

2. Prepare the dataset

Deblur-NeRF Synthetic Dataset (Re-rendered)

As described in the previous BAD-NeRF paper, we re-rendered Deblur-NeRF's synthetic dataset with 51 interpolations per blurry image.

Additionally, in the previous BAD-NeRF paper, we directly run COLMAP on blurry images only, with neither ground-truth camera intrinsics nor sharp novel-view images. We find this is quite challenging for COLMAP - it may fail to reconstruct the scene and we need to re-run COLMAP for serval times. To this end, we provided a new set of data, where we ran COLMAP with ground-truth camera intrinsics over both blurry and sharp novel-view images, named bad-nerf-gtK-colmap-nvs:

Download link

Deblur-NeRF Real Dataset

You can directly download the real_camera_motion_blur folder from Deblur-NeRF.

Your Custom Dataset

1. Use the ns-process-data tool from Nerfstudio to process deblur-nerf training images.

   For example, if the dataset from BAD-NeRF is in llff_data, execute:

   ns-process-data images \
       --data llff_data/blurtanabata/images \
       --output-dir data/my_data/blurtanabata
   

2. The folder data/my_data/blurtanabata is ready.

Note: Although nerfstudio does not model the NDC scene contraction for LLFF data, we found that scale_factor = 0.25 works well on LLFF datasets. If your data is captured in a LLFF fashion (i.e. forward-facing), instead of object-centric like Mip-NeRF 360, you can pass the scale_factor = 0.25 parameter to the nerfstudio dataparser (which is already set to default in our DeblurNerfDataParser), e.g., ns-train bad-nerfacto --data data/my_data/my_seq --vis viewer+tensorboard nerfstudio-data --scale_factor 0.25

3. Training

For Deblur-NeRF synthetic dataset and Deblur-NeRF real dataset, train with:

ns-train bad-nerfacto \
    --data data/bad-nerf-gtK-colmap-nvs/blurtanabata \
    --vis viewer+tensorboard \
    deblur-nerf-data

ns-train bad-nerfacto \
    --data data/real_camera_motion_blur/blurdecoration \
    --vis viewer+tensorboard \
    deblur-nerf-data

For custom data processed with ns-process-data, train with:

ns-train bad-nerfacto \
    --data data/my_data/blurtanabata \
    --vis viewer+tensorboard \
    nerfstudio-data --eval_mode "all"

4. Render videos

ns-render interpolate \
  --load-config outputs/tanabata/bad-nerfacto/<your_experiment_date_time>/config.yml \
  --pose-source train \
  --frame-rate 30 \
  --interpolation-steps 10 \
  --output-path renders/<your_filename>.mp4

Note1: You can add the --render-nearest-camera True option to compare with the blurry inputs, but it will slow down the rendering process significantly.

Note2: The working directory when executing this command must be the parent of outputs, i.e. the same directory when training.

Note3: You can find more information of this command in the nerfstudio docs.

5. Debug with your IDE

Open this repo with your IDE, create a configuration, and set the executing python script path to <nerfstudio_path>/nerfstudio/scripts/train.py, with the parameters above.

Evaluation

Image deblurring

Model	Dataset	PSNR↑	SSIM↑	LPIPS↓	Train Time (steps@time)
BAD-NeRF (paper)	Cozy2room	32.15	0.9170	0.0547	200k@11h
bad-nerfacto	Cozy2room	29.74 / 31.59	0.8983 / 0.9403	0.0910 / 0.0406	5k@200s / 30k@18min
BAD-NeRF (paper)	Factory	32.08	0.9105	0.1218	200k@11h
bad-nerfacto	Factory	31.00 / 32.97	0.9008 / 0.9381	0.1358 / 0.0929	5k@200s / 30k@18min
BAD-NeRF (paper)	Pool	33.36	0.8912	0.0802	200k@11h
bad-nerfacto	Pool	31.64 / 33.62	0.8554 / 0.9079	0.1250 / 0.0584	5k@200s / 30k@18min
BAD-NeRF (paper)	Tanabata	27.88	0.8642	0.1179	200k@11h
bad-nerfacto	Tanabata	26.88 / 29.32	0.8524 / 0.9133	0.1450 / 0.0895	5k@200s / 30k@18min
BAD-NeRF (paper)	Trolley	29.25	0.8892	0.0833	200k@11h
bad-nerfacto	Trolley	27.45 / 31.00	0.8675 / 0.9371	0.1222 / 0.0445	5k@200s / 30k@18min
BAD-NeRF (paper)	ArchViz-low	31.27	0.9005	0.1503	200k@11h
bad-nerfacto	ArchViz-low	26.70 / 27.03	0.8893 / 0.9046	0.1672 / 0.1267	5k@200s / 30k@18min
BAD-NeRF (paper)	ArchViz-high	28.07	0.8234	0.2460	200k@11h
bad-nerfacto	ArchViz-high	26.22 / 27.32	0.8649 / 0.8894	0.2504 / 0.2061	5k@200s / 30k@18min

Tested with AMD Ryzen 7950X CPU + NVIDIA RTX 4090 GPU, on Manjaro Linux, with CUDA 12.1 and PyTorch 2.0.1. Train speed may vary with different configurations.

Citation

If you find this useful, please consider citing:

@misc{zhao2023badnerfs,
    title     = {{Bad-RFs: Bundle-adjusted Radiance Fields from Degraded Images with Continuous-time Motion Models}},
    author    = {Zhao, Lingzhe and Wang, Peng and Liu, Peidong},
    year      = {2023},
    note      = {{https://github.com/WU-CVGL/Bad-RFs}}
}

@InProceedings{wang2023badnerf,
    title     = {{BAD-NeRF: Bundle Adjusted Deblur Neural Radiance Fields}},
    author    = {Wang, Peng and Zhao, Lingzhe and Ma, Ruijie and Liu, Peidong},
    month     = {June},
    year      = {2023},
    booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
    pages     = {4170-4179}
}

Acknowledgment

• Kudos to the Nerfstudio team for their amazing framework:

@inproceedings{nerfstudio,
	title        = {Nerfstudio: A Modular Framework for Neural Radiance Field Development},
	author       = {
		Tancik, Matthew and Weber, Ethan and Ng, Evonne and Li, Ruilong and Yi, Brent
		and Kerr, Justin and Wang, Terrance and Kristoffersen, Alexander and Austin,
		Jake and Salahi, Kamyar and Ahuja, Abhik and McAllister, David and Kanazawa,
		Angjoo
	},
	year         = 2023,
	booktitle    = {ACM SIGGRAPH 2023 Conference Proceedings},
	series       = {SIGGRAPH '23}
}