A minimalist pytorch implementation of: "Deep Geometric Functional Maps: Robust Feature Learning for Shape Correspondence" [1], appeared in CVPR 2020.
This implementation runs on python >= 3.7, use pip to install dependencies:
pip3 install -r requirements.txt
The preprocessing code will be added later. For the moment, we refer the reader to the original implementation of GeomFmaps to download the data and the preprocessing code.
It should be noted that for each dataset (faust, scape, etc), this module expect that the dataset folder contains 3 folders:
offfolder: this folder contains the meshesspectralfolder: this folder contains the laplace beltrami related data. It's composed from files having the same name as theofffolder. Each fileis a.matcontaning adictcontaining three keys:evals,evecsandevecs_trans. This files are created by the preprocessing code.corresfolder: this folder contains the ".vts" files necessary for the calculation of the ground truth maps.
Use the config.yaml file to specify the hyperparameters as well as the dataset to be used.
Use the train.py script to train the GeomFmaps model.
python3 train.py
[1] Deep Geometric Functional Maps: Robust Feature Learning for Shape Correspondence