M.Sc. or Ph.D. - UCL, Belgium.
Computer Vision and Deep Learning
Context: Democratic and personalized production of multimedia content is one of the most exciting challenges that content providers are facing today. In our project, we address this challenge by building on deep learning and computer vision tools to automate the collection and personalized distribution of audiovisual contents. In a typical application scenario, the acquisition set-up is composed of cameras, which, for example, cover a basket-ball field. Analysis and interpretation of the scene are exploited to decide what to show or not to show about the event, so as to produce a video composed of a valuable subset from the streams provided by each individual camera. The system provides a practical solution to cover local events at low cost, as no technical team or cameraman is involved anymore. It also promotes personalization of content, since knowledge about the content of the scene can be exploited to adapt and personalize content summarization to the individual user needs. The process involves numerous integrated technologies and methodologies, including but not limited to automatic scene analysis, camera viewpoint selection and control, and generation of summaries through automatic organization of stories.
Our project involves one academic partner (University of Liège), and gets content from two industrial actors (www.keemotion.com , a spin-off from our research group deploying autonomous sport production solutions, and www.evs.com, a world-leading company in broadcast and live sport production).
Status of the project: This 4-year project did start in October 2017. Datasets have been collected and annotated. Convolutional Neural Networks (CNNs) models have been developed and trained to detect objects of interest (players, ball, and referees) in sport scenes. Two PhD researchers are currently working on the improvement of those models in terms of accuracy and computational needs.
Open position: The open position is devoted to scene interpretation (e.g. team recognition, pose recognition for action recognition, player tracking) and/or to the prediction of the audience attention center (learning how to infer the visual attention in teamsport scenes). It implies the development of state-of-the-art and novel deep neural network architectures, as well as the implementation of training and testing pipelines to assess the performance of these architectures on relevant image processing tasks.
Applications should include a detailed resume, including grade sheets for B.Sc. and M.Sc.
Names and complete addresses of referees are welcome.
Please send applications by email to: email@example.com.