CSE802 Face Augmentation and Recognition Project

Dataset

We take CASIA-Webface[1] dataset as our training data for Face recognition model. The facial landmark aligned dataset is available from insightface

Preprocessing

Training data

We download the dataset from the above link insightface you might need some dependencies which can be installed with conda install --file requirements.txt

You can either prepare the training data yourself, or download it from our shared drive.

Prepare the data yourself.

cd preprocessing/training_data
unzip faces_webface_112x112.zip
python prepare_training_data.py

Or download from the link our shared drive

GAN Generated Images

We used the code provided at DiscoFaceGAN [2] to generate the augmented GAN images. The dependencies required to run the code are listed on their Github link provided under "Testing Requirements". The released paper for this work is also available at this link.

Adversarial Examples:

The code to generate the adversarial examples are in adversarial_examples.py script.

Reference

[1] Dong Yi, Zhen Lei, Shengcai Liao, Stan Z. Li. Learning Face Representation from Scratch. arXiv:1411.7923, 2014.

[2] Yu Deng, Jiaolong Yang, Dong Chen, Fang Wen, and Xin Tong. Disentangled and controllable face image generation via 3d imitative-contrastive learning. CoRR, abs/2004.11660, 2020.