Have you ever watched a video and seen an amazing outfit worn by a character, wanting to buy the same for yourself? Recommendations of items based on what we see is a well-known field. But what if these recommendations could be made dynamically based on the events happening in the video, recommending items of featured clothing while still maintaining the privacy of the user? This is our brand new featured clothing model that is part of Azure Video Indexer, a solution that generates insights from videos to help understand and search videos in media content and archives.

In this talk, I will explain how we can detect the main clothing items that are relevant to the viewer based on the changing content of a video, without the need of any personal data. Our model relies on multiple AI systems using both audio and visual domains, such as people detection, key moment detection, celebrity identification, and many more.