Huiqiang Sun¹, Xingyi Li¹, Liao Shen¹, Xinyi Ye¹, Ke Xian², Zhiguo Cao^1*,

¹School of AIA, Huazhong University of Science and Technology, ²School of EIC, Huazhong University of Science and Technology

Project | Paper | Video | Supp

This repository contains the official PyTorch implementation of our CVPR 2024 paper "DyBluRF: Dynamic Neural Radiance Fields from Blurry Monocular Video".

Installation

git clone https://github.com/huiqiang-sun/DyBluRF.git
cd DyBluRF
conda create -n dyblurf python=3.7
conda activate dyblurf
pip install -r requirements.txt

Dataset

The dataset consists of 6 dynamic scenes with motion blur. You can download this dataset from this link.

Each scene contains the following contents:

• images: blurry image sequence from left camera.
• images_xxx: resized blurry images from left camera.
• disp: depth map of the blurry images.
• flow_i1: optical flow of the blurry images.
• motion_masks: coarse motion mask of the blurry images.
• sharp_images: sharp image sequence from left camera.
• inference_images: sharp image sequence from right camera.
• poses_bounds.npy: camera poses of left blurry images computed by colmap.

Note: The camera parameters in poses_bounds.npy are arranged alternately for left and right cameras according to the time sequence of the video frames.

Training

python train.py --config configs/stereo_blur_dataset/xxx.txt

Citation

If you find our work useful in your research, please consider to cite our paper:

@article{sun2024_dyblurf,
    title={DyBluRF: Dynamic Neural Radiance Fields from Blurry Monocular Video},
    author={Sun, Huiqiang and Li, Xingyi and Shen, Liao and Ye, Xinyi and Xian, Ke and Cao, Zhiguo},
    journal={arXiv preprint arXiv:2403.10103},
    year={2024}
}