/DIP-GM

Mixing Integer Linear Programming and Deep Graph Matching

Primary LanguagePython

Deep graph matching meets mixed-integer linear programming: Relax at your own risk ?

This repository contains implementation of the paper: Deep graph matching meets mixed-integer linear programming: Relax at your own risk ?

avatar

The project directory is composed as follows:

  • experiments: contains configuration files to reproduce the results reported in the paper;
  • models: contains the implementation of the model based on difference graph matching solvers, such as DIP(ours), power_iteration and sinkhorn
  • utils: several utils;

Get started

Preliminary

  1. Check if findutils(>=4.7.0) is available
  2. Check if hdf5 is installed (apt install libhdf5-serial-dev)
  3. Check if cuda 10.1 and cudnn 7 are available
  4. Check if texlive-latex-extra is installed (apt install texlive-latex-extra)
  5. Check if torch_geometric is installed
  6. Check if gurobipy is installed

Download data sets

Run the following command to download data sets PascalVOC and SPair-71K

chmod +x ./download_data.sh && ./download_data.sh

Training

Run training and evaluation

python3 train_eval.py path/to/your/json

where path/to/your/json is the path to your configuration files. Our experimental configuration files are in ./experiments.