This code implements Multiscale Graph Convolutional Network (MGCN) in our SIGGRAPH 2020 paper:

"MGCN: Descriptor Learning using Multiscale GCNs"

by Yiqun Wang, Jing Ren, Dong-Ming Yan, Jianwei Guo, Xiaopeng Zhang, Peter Wonka.

Please consider citing the above paper if this code/program (or part of it) benefits your project.

	conda create -n MGCN python=3.7     # (options: 3.X)
	source activate MGCN                # (create and activate new environment if you use Anaconda)
	
	conda install pytorch=1.5.0 torchvision cudatoolkit=10.1 -c pytorch     # (options: 10.X)
	pip install torch-scatter==2.0.4+cu101 -f https://pytorch-geometric.com/whl/torch-1.5.0.html
	pip install torch-sparse==0.6.3+cu101 -f https://pytorch-geometric.com/whl/torch-1.5.0.html
	pip install torch-cluster==1.5.4+cu101 -f https://pytorch-geometric.com/whl/torch-1.5.0.html
	pip install torch-geometric==1.4.3

1. Put models in the folder MGCN/datasets/Faust/data_mesh/ for FAUST dataset or MGCN/datasets/Scape/data_mesh/ for SCAPE dataset

2. generate Wavelet Energy Decomposition Signature (WEDS) descriptors and wavelets using Repository WEDS or directly download processed data (traning.pt and test.pt) below and put them into the folder MGCN/datasets/Faust/processed/ or MGCN/datasets/Scape/processed/.

3. Training examples

	# for FAUST
	python MGCN_FAUST.py
	# for SCAPE
	python MGCN_SCAPE.py

4. Restore checkpoint and generate descriptors

	# for FAUST
	python MGCN_FAUST.py --uc --gd -l --ln=mgcn_faust-300
	# for SCAPE
	python MGCN_SCAPE.py --uc --gd -l --ln=mgcn_scape-300

5. Dense correspondence can be obtained by directly computing the nearest-neighbor using the L2 distance in descriptor space, and this Repository can be used to visualize the descriptors.

@article{wang2020mgcn,
  title=      {MGCN: Descriptor Learning using Multiscale GCNs},
  author=     {Wang, Yiqun and Ren, Jing and Yan, Dong-Ming and Guo, Jianwei and Zhang, Xiaopeng and Wonka, Peter},
  journal=    {ACM  Trans. on Graphics (Proc. {SIGGRAPH})},
  year=       {2020},
}

This program is free software; you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation; either version 2 of the License, or (at your option) any later version.