This repository contains my team's code submissions for the assignments in the course on Deep Learning, ELL888 at IIT Delhi.
The task is to build a neural network library from scratch. The network shall be used to classify letters in the Extended MNIST dataset. The library has support for dropout, batch normalization, L1 and L2 regularizations. It also supports use of different loss and activation functions. The best accuracy for a 9 class classification stands at 98.5%.
This time, we had to train a network to classify frames in a video based on the speaker. The training data consisted of YouTube videos of 6 speakers with different facial features. Faces were extracted using OpenCV's haar cascade and pretrained networks were used for face classification.