This app is a proof-of-concept for online retailers

I am a watch enthusiast who built the first version of this watch comparison prototype early 2016. This prototype aims to provide a proof-of-concept in relation to enhancing the online shopping experience of people visiting a watch site. More specifically, this app showcases an example when the visitor is allowed to search and compare watches based on design similarity and color scheme similarity. These are features that do not ship in regular shopping sites where searches typically are limited to brand, price, gender, etc. To build the watch database about 4000 images were scraped from an online retailer. To support the design-based tool I have used a convolutional neural network (CNN), therefore expect the algorithm to take about 15s to generate the results (using the caffe library from Berkeley running on CPU mode). As of now the prototype will NOT work on any random image of a watch but was built to work on a seller database. You can find images to query the prototype at : Sample pictures or instead can use our random image retrieval demo that is triggered by pressing the button in the front page (recommended as the upload feature crashes sometimes due to memory requirements of the CNN and the limited memory my free EC2 instance provides). Only jpg format supported so far. Want to know more? Fire me an email to: campanacue at gmail.com Thanks for Watching...

Copyright 2016 @MAISONlabs