This repository provides the implementation of our post-processing approach for floater removal in Neural Radiance Fields, specifically Instant-NGP.
A Post Processing Technique to Automatically Remove Floater Artifacts in Neural Radiance Fields
Tristan Wirth, Arne Rak, Volker Knauthe, Dieter W. Fellner
Computer Graphics Forum v42, 2023
A Dockerfile is provided for ease of use.
Python bindings for manipulating the density grid in Instant-NGP are provided in our Instant-NGP fork.
- NVIDIA GPU (RTX 20 Series and above)
- CUDA Toolkit (11.2+)
- Docker and Docker Compose
- For evaluation, you may download the Nerfbusters dataset and insert it into the dataset folder. It will be mounted in the docker container.
# build image
docker compose -f ./docker-compose.yml build floatyremoval
# optional if Instant-NGP GUI is to be used
xhost local:root
# open bash inside the container
docker compose -f ./docker-compose.yml run floatyremoval /bin/bash
From the Docker container's bash, run the colmap2nerf.py script to generate a transforms.json file for Instant-NGP.
cd dataset/pikachu
python3 /opt/instant-ngp/scripts/colmap2nerf.py --run_colmap --images images_2
Example usage of our post-processing approach for floater removal is shown in eval.py.
cd /volume
python3 eval.py dataset/pikachu
@article {10.1111:cgf.14977,
journal = {Computer Graphics Forum},
title = {{A Post Processing Technique to Automatically Remove Floater Artifacts in Neural Radiance Fields}},
author = {Wirth, Tristan and Rak, Arne and Knauthe, Volker and Fellner, Dieter W.},
year = {2023},
publisher = {The Eurographics Association and John Wiley & Sons Ltd.},
ISSN = {1467-8659},
DOI = {10.1111/cgf.14977}
}