Image-Based Face Classification and Verification
- 1 minDeep Learning Course Project at CMU.
This project aims to use Convolutional Neural Networks (CNNs) to design an end-to-end system for multi-label face classification and face verification.
My work:
- Implemented one recent baseline model ConvNet toward the design of a vision Transformer.
- Applied data augmentation techniques to images including random augmentation, random affine.
- Performed Triplet loss to model the similarities among triplets of faces on Face Verification tasks.
- Obtained over 90% in Face Classification tasks, and over 96% in Face Verification tasks. Ranked 14/354 on Kaggle Competition.
A typical Face Classification Architecture
Face Verification Architecture