https://github.com/ahmedbesbes/Understanding-deep-Convolutional-Neural-Networks-with-a-practical-use-case-in-Tensorflow-and-Keras

What makes convnets so powerful at image classification?
https://github.com/ahmedbesbes/Understanding-deep-Convolutional-Neural-Networks-with-a-practical-use-case-in-Tensorflow-and-Keras

article blog computer-vision convnet convolution-filter convolutional-neural-networks data-science dataset deep-learning deep-learning-tutorial image-classification kaggle kaggle-cats kdd keras keras-tutorials python tensorflow

Last synced: about 1 month ago
JSON representation

What makes convnets so powerful at image classification?

• Host: GitHub
• URL: https://github.com/ahmedbesbes/Understanding-deep-Convolutional-Neural-Networks-with-a-practical-use-case-in-Tensorflow-and-Keras
• Owner: ahmedbesbes
• Created: 2017-11-16T10:59:02.000Z (almost 7 years ago)
• Default Branch: master
• Last Pushed: 2017-11-21T08:58:44.000Z (almost 7 years ago)
• Last Synced: 2024-05-08T00:26:07.018Z (5 months ago)
• Topics: article, blog, computer-vision, convnet, convolution-filter, convolutional-neural-networks, data-science, dataset, deep-learning, deep-learning-tutorial, image-classification, kaggle, kaggle-cats, kdd, keras, keras-tutorials, python, tensorflow
• Language: Jupyter Notebook
• Homepage: https://ahmedbesbes.com/understanding-deep-convolutional-neural-networks-with-a-practical-use-case-in-tensorflow-and-keras.html
• Size: 9.62 MB
• Stars: 45
• Watchers: 5
• Forks: 31
• Open Issues: 0
• Metadata Files:
  • Readme: Readme.md

Awesome Lists containing this project

README

        
## 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:

```bash

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

    ```bash

    pip install graphviz  

    pip install pydot  

    ```