Use AI automated metadata generation to categorize content faster and more accurately

• Background

  A global photo app company that allows photographers to upload their photos came to Clarifai looking for a way to keep up with the demand for accurately tagging and rating an ever increasing volume of visual assets that was now averaging 5,000 photo uploads a day.

• Information

  Use Case

  Visual search

  Industry

  Software

  Client

  Mobile Phone App

Challenge

The company wanted to find a way to showcase their large volumes of user-uploaded images by rating each image on a scale of 1-10 and then promote their best assets on their website and through social media to boost sales. The company started out by using a team of five moderators to do this, but quickly found out that they couldn’t keep up with the workflow. They needed an automated solution that was highly-scalable, and quick-to-implement to categorize and rank the images that they received daily. Their outcome was to enhance user “findability” and make it convenient for customers to choose content based on image rankings. As a part of the process for tagging and ranking large asset volumes, they also needed a way to remove any ranking biases, such as removing the photo “popularity contest” element, e.g., neutralizing celebrity photos from getting the most “likes” even though their photos may not be great.

Solution

Before choosing Clarifai, the CEO and Co-Founder tested computer vision point solutions from Imagga, Microsoft and Google. The company selected Clarifai based on its superior accuracy, customer support, easy-to-follow documentation, intuitive platform tools, and multi-language support.

Clarifai’s deep learning AI and visual similarity search capabilities was used to first group similar photos by content and then directly compare a series of similar photos side-by-side to surface the best one by visual quality. The Clarifai pre-trained, out of the box celebrity model identified famous people in photos and was used to adjust the ranking algorithm in the company’s photo app. The Clarifai AI lifecycle platform was then used to automate the metadata generation process to accurately tag the large volume of images with the appropriate ranking.

Results

The company was able to build a solution using Clarifai’s visual similarity search and metadata generation capabilities in just three days and rolled out the application in just two weeks. The company was able speed metadata tagging by 100x over its human moderation team and is now able to rank all of its assets as they came in each day. They also improved tagging accuracy by 98% and increased shopper engagement by increasing average page duration time from one minute to four minutes. The company also doubled the amount of returning monthly website visitors within 2 months.

The company is planning to build additional use cases, including content moderation to ensure the appropriateness of its user submitted content. The company wants to have the ability to centralize and keep track of its multiple use cases using a single AI lifecycle platform and eliminate any need for additional toolsets and resources.

• Using Clarifai, we were able to create a sticky, gamified experience based on the ability for the user to directly compare photos of similar subject matter and choose the best one. Once a photographer uploads a photo, Clarifai’s API automatically recommends relevant and accurate categories and tags to the user.”

  CEO & Founder