Dog Breeds Classification with CNN Transfer Learning

Table of Contents

1. Installation
2. Project Overview
3. File Descriptions
4. Results
5. Licensing, Authors, and Acknowledgements

Installation

Beyond the Anaconda distribution of Python, the following packages need to be installed:

• opencv-python==3.2.0.6
• h5py==2.6.0
• matplotlib==2.0.0
• numpy==1.12.0
• scipy==0.18.1
• tqdm==4.11.2
• scikit-learn==0.18.1
• keras==2.0.2
• tensorflow==1.0.0 `

Project Overview

In this project, I built and trained a neural network model with CNN (Convolutional Neural Networks) transfer learning, using 8351 dog images of 133 breeds. CNN is a type of deep neural networks, which is commonly used to analyze image data. Typically, a CNN architecture consists of convolutional layers, activation function, pooling layers, fully connected layers and normalization layers. Transfer learning is a technique that allows a model developed for a task to be reused as the starting point for another task. The trained model can be used by a web or mobile application to process real-world, user-supplied images. Given an image of a dog, the algorithm will predict the breed of the dog. If an image of a human is supplied, the code will identify the most resembling dog breed.

File Descriptions

Below are main foleders/files for this project:

1. haarcascades
   • haarcascade_frontalface_alt.xml: a pre-trained face detector provided by OpenCV
2. bottleneck_features
   • DogVGG19Data.npz: pre-computed the bottleneck features for VGG-19 using dog image data including training, validation, and test
3. saved_models
   • VGG19_model.json: model architecture saved in a json file
   • weights.best.VGG19.hdf5: saved model weights with best validation loss
4. dog_app.ipynb: a notebook used to build and train the dog breeds classification model
5. extract_bottleneck_features.py: functions to compute bottleneck features given a tensor converted from an image
6. images: a few images to test the model manually

Note: The dog image dataset used by this project can be downloaded here: https://s3-us-west-1.amazonaws.com/udacity-aind/dog-project/dogImages.zip The human image dataset can be downloaded here: https://s3-us-west-1.amazonaws.com/udacity-aind/dog-project/lfw.zip

Results

1. The model was able to reach an accuracy of 72.97% on test data.
2. If a dog image is supplied, the model gives a prediction of the dog breed.
3. The model is also able to identify the most resembling dog breed of a person.

Project files can be found in this github repo: https://github.com/swang13/dog-breeds-classification More discussions can be found in this blog: https://medium.com/@wangshuocugb2005/dog-breeds-classification-with-cnn-transfer-learning-92217cba3129

Licensing, Authors, Acknowledgements

Credits must be given to Udacity for the starter codes and data images used by this project.