A Framework for Contextual Recommendations Using Instance Segmentation

Due to the restrictive measures to prevent COVID-19 from spreading, an increasingly large number of viewers are eschewing traditional television programs, resorting to streaming and on-demand platforms. This rapid change in audience preference, combined with the great appeal of streaming services, has constituted a form of "threat" for traditional advertising, causing advertisers and advertising agencies to adapt by participating in content that is, among others, supported by online advertising and streaming platforms. In this work, a novel framework for contextual recommendations using instance segmentation in movies is presented. The proposed service employs deep learning and computer vision algorithms to automatically detect objects in real-time on video streams. The experiments conducted offered satisfactory results regarding both the mAP (mean average precision) for the bounding box and the masks and the continuous decrease of the loss, as well as the correctly detected objects in real time.