keras_image_tutorial

Welcome to the tutorial on transfer learning for deep neural network image recognition models!

In this tutorial, you will learn

• What is the Google Inception model
• Transfer learning
• How to fine tune a transfer learning model

Installation

We recommend using the Docker container, but you can also install the required packages using Anaconda.

Docker

To start the tutorial, please clone this repository and set up the Docker container with the following steps:

1. Install Docker from https://www.docker.com/get-started. For linux, try sudo apt-get install docker.io

2. Run the Docker image:

   docker run -v "$(pwd)/notebook:/notebooks" -p 8888:8888 -p 6006:6006 docker.io/burk/champagne-image-tutorial
   

   • If you are using Docker for Windows, please first configure sharing as explained here. Then start the tutorial container by running

     docker run -v "(absolute path of the repository)/notebook:/notebooks" -p 8888:8888 -p 6006:6006 docker.io/burk/champagne-image-tutorial`
     

3. Copy the url given in the terminal window and open it in a browser

You can also build the Docker container yourself:

1. Build the Docker image: docker build . --tag tutorial:v1
2. Start the container: docker run -v "$(pwd)/notebook:/notebooks" -p 8888:8888 -p 6006:6006 tutorial:v1

Anaconda

If the Docker file doesn't work for you for some reason, please try to install the libraries needed using Anaconda:

1. Download and install Anaconda (https://www.anaconda.com/download/)
2. Create a Python 3.6 environment: conda create -n py36 python=3.6 anaconda
3. Switch to the new environment: activate py36 or source activate py36
4. Clone or download the tutorial material to your local computer
5. Open a notebook: jupyter notebook
   • If the notebook window doesn't open automatically in your browser, please copy the url given and open it in a browser
6. Run the first two cells in the keras_tutorial notebook from this repository to install the libraries needed

FAQ

• I got the error "docker: Error response from daemon: driver failed programming external connectivity on endpoint relaxed_varahamihira: Bind for 0.0.0.0:8888 failed: port is already allocated."

  • This means the port 8888 is already in use. It often occurs after you kill a notebook server. You can just pick a different port from the first port number in the docker run line, e.g. docker run -v "$(pwd)/notebook:/notebooks" -p 7777:8888 tutorial:v1

• Can I use Python 3.7?

  • As of 2018.12.03, Tensorflow hasn't released the official support for Python 3.7 yet, so we are using Python 3.6 for this tutorial at the moment.

• I got a FileNotFoundError when running cell 10!

  • We encourage you to find the pictures you like for testing, so you need to put some pictures there or change the image path called in the function.

• Can I use Linux subsystem in Windows?

  • It's quite tricky to make Docker work for the Linux subsystem in Windows, so we suggest to use a native environment. Or try the Anaconda option instead.