Masked-Unmasked Face Recognition

This project performs identity recognitions of masked faces. Our recognition accuracy was the highest among all groups choosing this project, and we received a Runner-Up among groups doing all projects. As the leader of a group of four, my contributions are as follows:

• Designed a synthesised mask generation algorithm based on ulib landmark detections and OpenCV
• Applied MtCNN for face localisation
• Finetuned a FaceNet and trained a classifier for masked face recognition at an accuracy of 80% on a 1000-identity subset of CelebA