Implemented a ResNet50 architecture from scratch using PyTorch (and Keras) to classify and verify facial images using a training set of over 1.6M images.
Achieved a test accuracy of over 74% and 92% on the classification and verification tasks respectively.
Datesets available on request.