ScatterNeRF: Seeing Through Fog with Physically-Based Inverse Neural Rendering (ICCV23)

Project Page | Paper

Official code implementation of ScatterNeRF. This work is being built on top of the great NeRF-Factory codebase.

Requirements

conda create -n nerf_factory -c anaconda python=3.8
conda install pytorch==1.11.0 torchvision==0.12.0 torchaudio==0.11.0 cudatoolkit=11.3 -c pytorch
pip3 install -r requirements.txt

## Or you could directly build from nerf_factory.yml
conda env create --file nerf_factory.yml

Dataset

We provide a ready-to-use sequence dataset sample, which can be downloaded here

Start the training

Both single and multiple GPU supported

python3 -m run --ginc configs/[model]/[data].gin
# ex) CUDA_VISIBLE_DEVICES=1,2,3 python3 run.py --ginc configs/scatternerf/tnt.gin --scene_name Sequence00_left_right

Render results

(Currently, only single-gpu supported)

python3 run.py --ginc configs/[model]/[data].gin --scene [scene] --ginb run.run_train=False
# ex) CUDA_VISIBLE_DEVICES=0 python3 run.py --ginc configs/scatternerf/tnt.gin --scene_name Sequence00_left_right --ginb run.run_train=False

License

Copyright (c) 2022 POSTECH, KAIST, and Kakao Brain Corp. All Rights Reserved. Licensed under the Apache License, Version 2.0 (see LICENSE for details)