This repository contains files for my deep learning project solutions for Projects 1-4 and the final project.
Title: Predicting Race, Age and Gender from Face: A Small-sample Example with Encoding
Abstract
Classifying facial images into appropriate age, gender and class groups is a practical problem in deep learning. Using FairFace dataset, we construct a two-stage method for classification with small dataset while reducing bias in them. In our two-stage method, we initially train a Variational Autoencoder (VAE) to project images to a latent dimension. Then, use the encodings from the latent dimension to train a single Convolutional Neural Network (CNN) classifier model with three outputs for classifying race, age and gender. As a point of reference, we employ a popular machine learning algorithm, Random Forest as baseline. We find the random forest model to be not good at the task.