This repo contains a demo of transfer-learning using Inception v3 and TensorFlow in a Jupyter notebook.
Much of the material is based on Google's Codelab TensorFlow for Poets
** If you don't want to set up the project dependencies yourself**, you can view results used in a presentation, with different humanoid species of the Star Trek universe here.
Clone this repo via
git clone --recursive git://github.com/foo/bar.git
or if you've already downloaded it, you need to init submodules manually:
cd <repo>
git submodule update --init --recursive
Install the Miniconda (or Anaconda) Python distribution. The code has been tested on Python 2.
Set up conda environment via conda env create -n transfer-learning -f environment.yml
The dataset-folder contains two folders: "Train" should contain subfolders named after the class of images they contain; "Test" should contain test images from multiple classes (not present in the training data).
You will need to build your dataset;
I just used Fatkun Batch Download Image to handle batch downloading from Google image search results.
Run the notebook:
# Activate our Python environment
source activate transfer-learning
# Start Jupyter and open the notebook
jupyter notebook