Explore GEN family of websites, social media experiences, blogs and more.

Unleashing ideas.
Strengthening startups.

Back to search results
    Date submitted
  • 31-Jan-2014

Frenzy

Abstract

Automated Visual Product Recognition.

Angela is a 22 year old fashionista who reads Instyle.com and fashion blogs to decide what clothing she wants to buy. She spots a sweater worn by Emma Stone. How does she find the sweater without having to spend hours searching retail stores?

Frenzy is a site plugin that identifies and tags clothing in web images automatically. Once installed, it scans each web page, takes inventory of the exact items, and transforms pictures into visual storefronts where users can buy clothing they spot in media. Site owners and app makers like InStyle will use Frenzy to automate the process of monetizing photos. It saves their staff from one hour to several days for each post; identifying products, searching for matched product links, managing payment programs, and updating expired links.

Vision APIs right now achieve less than 1% accuracy with fashion products and are too unreliable to scale. We have a working algorithm that identifies clothing with 65% accuracy.

Video

Original YouTube URL: Open

Introduction Video

Photos

Additional Questions

Who is your customer?

We are introducing Frenzy to popular style blogs and digital magazine segments which accounts for over 23 billion monthly site visits. In the long-term we want to partner with big media companies so we can integrate Frenzy technology with video, smartdevices, shopping channels, advertisements and social networks. In the next phase we plan to classify products in more categories like beauty, health & wellness, electronics and home décor. Frenzy’s mission is to enable automated commerce for every product in a consumer’s field of vision with perfect accuracy.

What problem does this idea/product solve or what market need does it serve?

All of the solutions to monetize content that exist today are manual. Frenzy is an automated solution that lets media sites earn money from their photographs at scale. Site owners install the plugin one time and continue to create content while the algorithm runs its course; showing exact and similar styles that appear in images for audiences to shop. Frenzy is able to parse through each item in any web image to determine the exact brand, SKU and where it can be purchased.

What attributes will make this idea/product successful? Why do you believe that those features will create success?

Vision APIs right now won’t classify fashion products accurately, nor do they offer a scalable solution for monetizing images. Frenzy’s proprietary language processing, image recognition, and web crawling models run seamlessly to achieve a 65% hit rate with fashion products while guiding users from inspiration to transaction. The system can already identify clothing seen in images with an accuracy that leaves humans in the dust – we had a face-off with expert celebrity fashion stylist Ali Levine (alilevine.com) and Frenzy outperformed her in every instance: speed, accuracy (# of exact matches) and efficiency (# of steps). Ali asked to join our beta once she realized how much more time she would have to photograph, style and post on her blog without the pain of monetizing content.

Explain how you (your team) will execute to make this idea/product successful? What gives you (your team) an advantage over others already in the market or new to this market?

Fashion on social media and style blogs, which exploded in popularity over the last 7 years, are giving us all the training data required to make automatic clothing recognition a reality. We use machine learning to give us an edge over competitors. Frenzy’s accuracy will rise to the 90% threshold by 2018 as we add more data in the pipeline. Each article processed widens the gap between Frenzy and competitors as the system learns to recognize specific brand name products in images so it won’t need to rely on software patents. Our sales team is in the process of securing exclusive content agreements with our site partners to prevent them from sharing training data with potential competitors.