Dog Breed Classifier Project

The motivation behind this project is to build a classification algorithm that predicts the breed of a given dog.The classification algorithm accepts a file path to an image and first determines whether the image contains a human, dog, or neither. Then,

• if a dog is detected in the image, return the predicted breed.
• if a human is detected in the image, return the resembling dog breed.
• if neither is detected in the image, provide output that indicates an error.

This project is also featured on the Medium website @ link

Table of contents

• Data and Code
• Prerequisites
• Instructions on running the application

Summary of the results

The algorithm uses a pre-trained InceptionV3 CNN to predict the image of a given dog within a reasonable level of accuracy.The model achieves an accuracy of around 78% in classifying 133 dog breeds. When given an image path the algorithm returns results in this format:

• 
• 

Below is an architecture of the custom CNN that was developed during the project.

Data and Code

• The main application is contained in the dog_app.ipynb jupyter notebook. All the functions and code, as well as the rationale behind decisions taken is contained in this notebook.
• Test images are in the images folder in the root directory.
• The main data set was provided by Udacity and can be accessed here:
  • Dog dataset
  • Human_dataset

Prerequisites

• Numpy
• Sklearn
• Sqlalchemy
• Pickle
• keras
• cv2
• glob
• seaborn
• os
• Matplotlib
• Pillow Jupyter notebook and python 3.6 are needed to run the notebooks and python scripts.

Instructions on running the application

1. Download the required data sets and if required modify the directory paths.
2. Access to a GPU is essential in training the models.