Face Recognition Using Subspace Methods

An example of predicted matches found by the algorithm

Similarity scores of the matches

Accuracy of the implemented methods

Method	Top 1 accuracy	Top 5 accuracy
MSM (baseline)	91.48	93.61
KMSM	80.85	91.48
Hybrid	91.48	93.61

Overview

Subspace based methods are implemented and evaulated using the YTC dataset.

1. Mutual Subspace Method
2. Kernel Mutual Subspace Method
3. Kernel Mutual Subspace Method + Constrained Mutual Subspace Method

Requirement

CuPy is used for linalg operataions.

conda create -n facerecog python=3.7
conda activate facrerecog
conda install cupy ConfigArgParse matplotlib

Usage

To run tests using configurations specified in config file

python run.py -c config/exp.conf

Features

TODO

Reference

1. 部分空間法の面白さに魅せられて, 福井 和広
2. 制約相互部分空間法と直交相互部分空間法の比較, 福井 和広 山口 修
3. LECTURE: KERNEL PCA, Rita Osadchy

Author

Pragyan Shrestha
University of Tsukuba, Empowerment Informatics 1st Year