Dog breed classification project

This is the repository of my capstone project for the Udacity Machine Learning engineer nanodegree. The code was developed in the dog_app jupyter notebook and the models was trained in a workspace provided by Udacity.

The main goal is to build an algorithm that takes an image path and if it’s a human image, it will predict the resembling dog breed. If there is a dog in the image, it will predict its breed.

Dataset

Dog dataset

Faces dataset

Steps

• 1. Exploratory data analysis.
• 2. Built a face detector with a haar cascade classifier.
• 3. Built a dog detector with a VGG-16 pre-trained model.
• 4. Create a CNN model from scratch, train it and test it.
• 5. Use transfer learning with VGG-16 model. Train and test it.
• 6. Built a final algorithm whith the human detector, dog detector and the last transfer learning model.