Understanding deep Convolutional Neural Networks with a practical use-case in Tensorflow and Keras

This jupyter notebook reassembles the code of this article

It also contains a trained CNN model, so that you can use it yourself and test it.

Summary of this article

This post is a 4-part tutorial where I:

1. Present an image dataset from the Cat vs. Dog Kaggle competition and explain the complexity of the image classification task
2. Go over the details about Convolutional Neural Nets, explaining their inner meachanisms and the reason why they perform better than fully connected networks.
3. Set up a deep learning dedicated environment on a powerful GPU-based EC2 instance from Amazon Web Services (AWS)
4. Train two deep learning models: one from scratch in an end-to-end pipeline using Keras and Tensorflow, and another one by using a pre-trained network on a large dataset.

These 4 parts are independent.

If you're looking to understand the theory behind convnets please refer to the article link posted above.

Environment setup:

1. Use Python 3.6: No hassle, intall the Conda distribution that encapsulates the PyData stack (SciPy, Pandas, Matplotlib, etc.). Here's the installation link

2. Install the lastest version of Tensorflow: https://www.tensorflow.org/install/install_windows I used a windows machine, same applies for Linux or Mac OS X

3. Install the following python dependencies:

pip install keras
pip install tqdm
pip install keras-tqdm
conda install -c conda-forge opencv 

4. [Optional] Dependencies to obtain GraphViz plots of the CNN architectures:

   1. Install Graphiz: http://www.graphviz.org/Download..php
   2. Add the Graphiz binaries to you PATH
   3. Install Graphiz Python bindings

   pip install graphviz  
   pip install pydot