Dog-Breed-Classifier

Project for Deep Learning Nanodegree, unit 3.

This project consists of developing an algorithm that could be used as part of a mobile or web app. The main purpose is to classify dog breeds. If a dog is detected in the image, it will provide an estimate of the dog's breed. If a human is detected, it will provide an estimate of the dog breed that is most resembling.

Steps:

1. Import Datasets
2. Detect Humans
3. Detect Dogs
4. Create a CNN to Classify Dog Breeds (from Scratch)
5. Create a CNN to Classify Dog Breeds (using Transfer Learning)
6. Write a simple algorithm
7. Test the algorithm

Results:

1. Using pure CNN to classify dog breeds has a test accuracy of 13%.
2. Creating a CNN using Transfer Learning (i.e. resnet101 architecture) generates 83% accuracy.