Linearized Partial Gromov Wasserstein

Installation

Example code for linear partial Gromov Wasserstein solver. The code reproduces the all the numerical examples in the paper.

Required packages

We suggest to install Pytorch, PythonOT, scipy, numba, sk-learning.

Can create necessary Conda environment with

conda env create -f environment.yml

Outline of repository

lib/ contains code of partial GW solvers and GW-based methods for all experiments. See lib/README.md for references.

shape_retrieval_3d/ contains code for shape retrieval experiment on 3D dataset. See shape_retrieval_3d/README.md for details. shape_retrieval_2d contains code for shape retrieval experiment on 2D dataset. See shape_retrieval_2d/README.md for details

mnist contains code for classification experiment in MNIST dataset. See mnist/READMe.md for details

citation

• In lib\HK, the code (LUOT/UOT solver) is imported from https://github.com/bernhard-schmitzer/UnbalancedLOT.
• In lib\mpgw, the code (mpgw sovler) is imported from https://github.com/lchapel/partial-GW-for-PU
• In ellipses\, the data and code (for visulization) in utils.py is imported from https://github.com/Gorgotha/LGW
• In shape_retrieval_2d, we import and modify the code (for 2D data generation and SVM classification) in https://github.com/Gorgotha/LGW.
• In shape_retrieval_3d, we import and modify the code (for SVM classification) in https://github.com/Gorgotha/LGW.