https://www.youtube.com/watch?v=QfNvhPx5Px8

1. Install docker
   http://docs.aws.amazon.com/AmazonECS/latest/developerguide/docker-basics.html#install_docker

2. Setup docker
   docker run -it gcr.io/tensorflow/tensorflow:latest-devel

3. Download images (use Fatkun batch download extension on google) Put images in this format: tf_files/star_wars/vader, tf_files/star_wars/darth_maul

4. Connect image files to docker (need at least 2 classes: tf_files/star_wars/vader, tf_files/star_wars/darth_maul)
   docker run -it -v /home/ec2-user/app/image_recognition_transfer_learning/tf_files:/tf_files/ gcr.io/tensorflow/tensorflow:latest-devel

5. Retraining

python tensorflow/examples/image_retraining/retrain.py \  
--bottleneck_dir=/tf_files/bottlenecks \  
--how_many_training_steps 500 \  
--model_dir=/tf_files/inception \  
--output_graph=/tf_files/retrained_graph.pb \  
--output_labels=/tf_files/retrained_labels.txt \  
--image_dir=/tf_files/star_wars

5. Create label_image.py (load image and graph to tensorflow, get predictions)
6. Use python label_image.py <image_path.jpg> command to predict image