-webkit-filter: blur(0);

Success story: How Microwork's annotations are fuelling the future of search

About Markable

Markable build smart AI-powered tools for e-commerce companies and publishers. In their own words they are 'bridging the gap between digital media content and e-commerce - making visual search a reality on every digital photo and video across any platform'. A use case could be Netflix. Imagine if pausing the screen on Netflix would bring up a window which displayed all the items of clothing currently onscreen with links to where you could purchase them online. It would be a case of relevant, unintrusive advertising.

The Problem

In Markable's mission to build smart tools for a more personalized online experience, they boast a 0.5 second discovery time and high accuracy. That's 0.5 seconds for an algorithm to recognise what's in the image, search a database for similar images, and present those images and relevant links to the user!

Plus they have very specific requirements for what the algorithm should recognise in an image. If the visual search is of a red velvet dress, it's not effective if the search results bring up dresses of different colours or patterns. So Markable need high-quality image annotation with very specific attributes. Enter Microwork...

The Solution

When our relationship with Markable began, our annotation tool didn't have all the features required to complete the task. We could annotate what article of clothing was in the image, but we were unable to get more specific than that. So we developed our tool on the job to meet Markable's needs. We added the ability to annotate any specific attributes of the clothing too - i.e. the type of fastener a jacket has, what type of pattern is on a dress or what type of shoes are being worn. These attributes allow us to create data that's capable of the specificity that Markable's search algorithms require.

Working with Markable, we also discovered other features that would be useful for meeting their requirements. At Markable's request, we adapted our tools to allow more specific filtering capable of handling diverse and high-volumes of data. In addition to this, we wanted to streamline the feedback process. We also added a feedback feature directly into our app to create a quicker and more efficient feedback loop in addition to our weekly meetings.

With Markable, we were able to develop our tool and process in order to meet their needs and help support their mission to enhance visual search. We can't wait to see the success they have in their mission.