README.md

HodgeNet | Webpage | Paper | Video

HodgeNet: Learning Spectral Geometry on Triangle Meshes
Dmitriy Smirnov, Justin Solomon
SIGGRAPH 2021

Set-up

To install the necessary dependencies, run:

conda env create -f environment.yml
conda activate HodgeNet

Training

To train the segmentation model, first download the Shape COSEG dataset. Then, run:

python train_segmentation.py --out out_dir --mesh_path path_to_meshes --seg_path path_to_segs

To train the classification model, first download the SHREC 2011 dataset:

wget -O shrec.tar.gz https://www.dropbox.com/s/4z4v1x30jsy0uoh/shrec.tar.gz?dl=0
tar -xvf shrec.tar.gz -C data

Then, run:

python train_classification.py --out out_dir

To train the dihedral angle stress test model, run:

python train_origami.py --out out_dir

To monitor the training, launch a TensorBoard instance with --logdir out_dir

To finetune a model, add the flag --fine_tune to the above training commands.

BibTeX

@article{smirnov2021hodgenet,
  title={{HodgeNet}: Learning Spectral Geometry on Triangle Meshes},
  author={Smirnov, Dmitriy and Solomon, Justin},
  year={2021},
  journal={SIGGRAPH}
}