Deep Learning projects

This repository contains rojects for Udacity's Artificial Intelligence Nanodegree term 2.

Environment Setup

The projects are run in an environment with Python 3 and all necessary packages installed. The environment can be created and configured by following the steps below:

1. conda env create -f aind-environment-osx.yml to create the environment (for OSX)
2. source activate aind to enter environment
3. pip install git+https://github.com/hmmlearn/hmmlearn.git to install the development version of hmmlearn 0.2.1

Project 1: Build Dog Breed Classification App

In this project, CNN models are created to classify dog breeds. Given an image of a dog, the algorithms will identify an estimate of the dog’s breed. If supplied an image of a human, the models will identify the resembling dog breed.

CNN models built from scratch as well as from transfer learning are compared. For CNN transfer learning, two different pre-trained models serving as the fixed feature extractor are compared. The CNN model with ResNet-50 achieved a test accuracy over 79% whereas the one with VGG-16 obtains a test accuracy around 46% after 20 epochs.