Machine Learning

Machine Learning2018-08-28T14:56:32+00:00

How do you tame a tsunami of content?

Massive amounts of media are produced on a daily basis – 300 hours of video are uploaded to YouTube every minute! And even if your content production is only a fraction of that, it can still be overwhelming to manage. Add interactivity to the mix and the challenge becomes even harder. Timeliness and accuracy are vital – you want to direct traffic to the right content at the right time.

So how can help you with this challenging task? We believe in working smarter, not harder. We harness machine learning & AI algorithms to efficiently enrich your media, during each step of the process. Read on to learn how we can help you – we’ll show you how it works with a fragment from the hit TV show New Girl.

Content selection

Content selection is mostly an issue with videos, where each video is made up out of thousands of frames. A selection needs to be made of frames that will be made interactive or shoppable so we can focus on the content with the most potential.

Instead of manually hitting the play & pause buttons over and over, we apply machine learning to the task. Our proprietary algorithm automatically selects frames with high image quality that contain the most items to be made interactive. This way we can efficiently focus our efforts on the parts of the video where they will have the most impact.

Image processing

Each frame of a 1080p HD video contains 2.1 mega pixels – that’s 2.1 million pixels! Again, we deploy machine learning algorithms, which scour visuals for elements with the potential to be interactive or shoppable. The system automatically detects faces and clothing, so we can ignore all of the irrelevant background elements.

Product recognition

We now have a selection of items that we want to make interactive or shoppable. But how do we make sure that we link to the right product each time? With the vast selection of products for sale, finding the right match is like searching for a needle in a haystack.

So again we turn to our robot friends for help, who match the items that we identified against a database of more than 380 million products. Do you want to offer affordable alternatives to inspirational haute couture outfits? Our system automatically detects the options that are the best match.

We can also provide constraints to guide our algorithm, for example telling it to only search for matches within the catalog of your brand or the brands that you partner with. This increases accuracy while reducing the workload by reducing the set of products to match against.

Want to get more inspiration from Jess Day and her friends? Check out the rest of our New Girl content!

The human element

During each step along the path from raw visual input to processed interactive content, human supervision provides quality control. Because the repetitive and labor-intensive tasks are automated, we can focus the human efforts on providing the best quality content possible, without ballooning the workload (and costs).

Our algorithms are intelligent and adaptive, which means that they learn from their mistakes and become even more accurate over time. If the system misidentifies a person or product and is corrected by human supervision, the algorithm will automatically adapt itself to avoid this from happening again in the future. The perfect marriage of man & machine!

“Thanks to our partnerships with leading research institutions like imec and the Universities of Ghent & Leuven we utilize cutting-edge machine learning and AI algorithms. Because we can efficiently process massive amounts of media, we deliver results faster, and at lower prices.”

Dimitri Poppe, Head of IT at

Are you ready to get (Artificially) Intelligent?

Contact us