Recolorable Posterization of Volumetric Radiance Fields

This is the official code release for the paper:

Recolorable Posterization of Volumetric Radiance Fields Using Visibility-Weighted Palette Extraction
Kenji Tojo and Nobuyuki Umetani
In Eurographics Symposium on Rendering (EGSR) 2022
[Project page]

Code structure

rgbsg contains the RGB-space convex hull analysis for palette extraction.

plenoctree contains the C++/CUDA library for radiance sampling, sample filtering, and stylized rendering.

For details, see the README.md under these directories.

Installation

Clone this repository recursively

git clone --recursive https://github.com/kenji-tojo/posternerf.git

Requirements

C++17, CUDA, and Python.
We tested on CUDA 11.4 and Python 3.8.

Please also setup the Python dependencies through

conda create -n posternerf python=3.8
conda activate posternerf
pip install --upgrade pip
pip install -r ./requirements.txt

Building

Please build the C++/CUDA library under plenoctree to run this demo.

Linux

mkdir build && cd build
cmake ..
make -j9

Windows

mkdir build && cd build
cmake .. -G "Visual Studio 16 2019"
cmake --build . --config Release

Data

Our method takes a pre-trained PlenOctree model as input.

To be self-contained, we release the models used in our experimetns here.
The original data can be found here.

You may put these resource files under the directory named data. Git will ignore them.

Palette extraction

Run the palette extraction via

python palette_extraction.py data/<model_name>.npz

The output will be saved to the output/<model_name> directory.

Or, you may download the palettes computed on our machine here.

Stylized rendering

To launch our run-time posterization and recoloring demo, run

python viewer.py data/<model_name>.npz --palette output/<model_name>/palette.npz

Please use the GUI to controll the stylization, such as enabling/disabiling the posteriation and performing recoloring.