oneshotlearning

This repository is the required code for the project "Applied Framework for One-shot Face Recognition using Siamese Neural Networks" for the Toronto Metropilitan University course EE8223: Deep Learning with Dr. Illanko.

The project uses a deep siamese neural network pair to perform face recognition. We developed a GUI to allow for tunable hyper-parameters. We train the network on two datasets, Labelled Faces in the Wild (LFW), and Youtube Faces (YTF). The data setup can be found below.

• Note: latest results have not been uploaded to github yet. Do not use the "results.csv" file as a reference.

Quick Start

Clone this repository, using

1. git clone https://github.com/husseinalijaafar/oneshotlearning
2. cd to oneshotlearning/ and run
3. pip install -r requirements.txt
4. Set up the data as shown below
5. run: python3 MainWindow.py
6. Select your required configuration, change layer sizes, activation functions etc.
7. Click "Initialize Network"

Data set up

Download the following dataset:

• LFW
• LFW_train.txt
• LFW_train.txt
• YTF

Extract and arrange the data in the following setups

The dataset folder should appear as

dataset
  lfw2
    lfw2
      ...
    splits
      train.txt
      test.txt
    weights
      ...
  ytf
    ytf_split
      test
      train
      val
      test.txt
      train.txt

For questions or concerns, please open an Issue on GitHub.

Credits

• Original paper is Gregory Koch, Richard Zemel, Ruslan Salakhutdinov
• Some of the network code used is from [Nevoit](https://github.com/nevoit/Siamese-Neural-Networks-for-One-shot-Image-Recognition
• Datasets: LFW YTF