This is a tool to visualize the similarities captured by deep learning features on small datasets of up to ~1000 images.

I finished this project in December 2015 as the final project of the CSPC 547 -- Infoviz course taught by Tamara Munzner at the University of British Columbia.

I also wrote a paper-like final report that explains the design choices behind this tool.

• /python contains code for data gathering (i.e., downloading images from twitter using tweepy) and computing deep features using Caffe.
• /js contains the bulk of the javascript code that runs the interface.

To compile the code install the dependencies and generate the bundle using browserify:

cd js/
npm install -g browserify
npm install -i static-kdtree
browserify main.js -o bundle.js

The code works out of the box for a dataset compiled from the twitter account Faces in things. If you want to use your own dataset you will have to generate the low-dimensional with a working implementation of t-SNE.

MIT