As part of our first meeting for the 2017-2018 year, we showed the following live demos. The main purpose was to show people the different advancements made in machine learning over the past few years.
-
Convolutional neural networks - Showed how to load in a trained CNN model in Keras and predict the class category of the image.
-
Fast Style Transfer - Code taken from this repository. Showed how to take any image, and add styles from famous paintings to those photographs.
-
Character RNN - Code taken from this repository. Showed how you can take a piece of Shakespeare text, and train a recurrent neural network to output similar looking text. Excellent blog post on how this is done.
In order to run these demos, you'll need TensorFlow and Jupyter.
The first step is to clone the repository.
git clone https://github.com/uclaacmai/Live-Demos.gitTo see the convolutional neural networks demo, just launch a Jupyter notebook by entering
jupyter notebook
To see style transfer (currently works with TF version 0.11 but check out the original repo for updates), run the following shell script with the argument being the style you want for the images. The choices of styles can be found here. The new images will be the resulting_images folder.
./styleTransfer.sh scream
To see character RNN (currently works with TF version 1.0), run the following shell script, and then go ahead and view the created text in shakespeareText.txt
./charRNN.sh