Multi-Object Matching (for Python)

Erebus Oh, Professor Stephen Phillips

Contact [email protected] for questions.

This repository is a Python implementation of Multi-Object Matching. To view their README, see Multi-ObjectMatchingREADME.md.

Setup

Anaconda Environment

Set up and activate the Anaconda Environment by running the command:

conda env create --file environment.yml
conda activate multiway

Pip Virtualenv Environment

Or set up with pip virtual env:

virtualenv env
source env/bin/activate
pip3 install -r requirements.txt

Dataset

We use the WILLOW-ObjectClass Dataset and Alexnet. Follow the steps below:

1. Download WILLOW-ObjectClass Dataset at dataset/

cd dataset/
wget http://www.di.ens.fr/willow/research/graphlearning/WILLOW-ObjectClass_dataset.zip
unzip WILLOW-ObjectClass_dataset.zip
# remove problematic image and annotation
rm -f WILLOW-ObjectClass/Face/image_0160.*
# there is an annotation error in Cars_030a.mat (coordinate swap between 6th and 7th keypoint), replace it with the correct one
mv Cars_030a.mat WILLOW-ObjectClass/Car/

2. Download Alexnet Weights at hypercols/, and then extract feature descriptor hypercolumn from AlexNet.

cd ../hypercols/
wget http://www.cs.toronto.edu/~guerzhoy/tf_alexnet/bvlc_alexnet.npy
python run_willow.py

Running the code

To run the code, type the command:

python test_willow