Dynamic Neural Radiance Field From Defocused Monocular Video

Xianrui Luo¹, Huiqiang Sun¹, Juewen Peng², Zhiguo Cao¹

¹Huazhong University of Science and Technology, ²Nanyang Technological University

Data | Paper

This repository is the official PyTorch implementation of the ECCV 2024 paper "Dynamic Neural Radiance Field From Defocused Monocular Video".

Usage

Dependency

• numpy
• scikit-image==0.21.0
• imageio
• configargparse
• opencv-python
• kornia
• torch
• torchvision

Training

python run_nerf.py --config configs/xxx.txt

We use the left view for training and the right view for evaluation.

python run_nerf.py --config configs/xxx.txt --render_test

The checkpoints are in Google Drive.

There are a total of 8 scenes used in the paper. You can download all the data in here.

Acknowledge

This source code is derived from Deblur-NeRF and NSFF. We appreciate the effort of the contributors to these repositories.