/PICT_smoke

[SIGGRAPH 2024] Official code for Physics-Informed Learning of Characteristic Trajectories for Smoke Reconstruction

Primary LanguagePython

Physics-Informed Learning of Characteristic Trajectories for Smoke Reconstruction

[Project] [Paper] [arXiv] [Video]

Yiming Wang, Siyu Tang, Mengyu Chu

SIGGRAPH 2024

Setup


# build environment with python 3.7
conda create -n pinf python=3.7
conda activate pinf 

# if ffmpeg is not installed (test by ffmpeg -version)
conda install -c conda-forge ffmpeg 
conda install ffmpeg

# requirments
pip install -r requirments

# raymarching
cd raymarching
pip install -e .

# test environment
python env_test.py

The dataset used in the paper can be downloaded from Goolge Drive.

Run

Training

Take the Cylinder scene as an example:

python train.py --config configs/cyl.txt

Testing

# velocity voxel output
python test.py --config configs/cyl.txt --testskip 1 --output_voxel --full_vol_output

# render novel view
python test.py --config configs/cyl.txt --render_only

# static object mesh
python test.py --config configs/cyl.txt --mesh_only

Installing problem

  • Ninja is required to load C++ extensions
pip install Ninja

Citation

@inproceedings{Wang2024PICT,
  author = {Wang, Yiming and Tang, Siyu and Chu, Mengyu},
  title = {Physics-Informed Learning of Characteristic Trajectories for Smoke Reconstruction},
  year = {2024},
  url = {https://doi.org/10.1145/3641519.3657483},
  doi = {10.1145/3641519.3657483},
  booktitle = {ACM SIGGRAPH 2024 Conference Papers},
  articleno = {53},
  numpages = {11},
  series = {SIGGRAPH '24}
}