Skip to content

Latest commit

 

History

History
38 lines (21 loc) · 1.81 KB

README.md

File metadata and controls

38 lines (21 loc) · 1.81 KB

I_LOVE_IKEA_SHINY_APP

This is the Ikea hackaton R Shiny app that we (I, Jos van Dongen and Thomas van Dongen) created on December 7,8,9 of 2017. The instructions for the app are quite simple, see images below. Our 3 minute pitch in pdf is also in this repo.

The app helps IKEA customers to look for Ikea products by means of image search. Take a picture, upload it and you will get the top matching IKEA products back.

So what steps are taken to get this app working?

  1. Scrape IKEA website using R package rvest for product images, in total we have around 9000 scraped IKEA images. The R code to scrape Ikea is also available on GitHub, see me ikeaScraper library.

  2. Use R Keras to put all images through a pretrained VGG16 network from which the top layers are removed. Each image is now a tensor, that we flatten to a vector (25.088 dimensional vector). So we end up with a 9000 by 25088 matrix.

  1. For a new image: put it trough the same network, and flatten this to a 25088 dimensional vector as well. Now determine for that image the top 10 closest IKEA images, using for example cosine similarity.

  2. Present the reults in a Shiny App.

  3. A live running shiny app can be found here, which I will probably close down by the end of 2017 because it costs me some money to keep up a droplet on Digital Ocean :-)