Marta Mrak
Lead R&D Engineer, BBC
Learning to Predict Pixels for Content Enhancement and Delivery
Monday, October 19th 10:30 AM – 12:00 AM (GMT-3)
Abstract
Machine learning techniques for more efficient video compression and enhancement have been developed thanks to breakthroughs in deep learning. These new techniques bring previously unforeseen capabilities. For example, adding colour to pixels of black and white video has traditionally been an expensive and time-consuming task – until now. Recent advances, such as availability of Generative Adversarial Networks (GANs) which are addressed in this talk, enabled the development of new colourisation algorithms which generate plausible colourful content.
However, such prediction of colour and pixels in general typically comes in the form of resource-hungry black-boxes, i.e. sometimes overly complex algorithms with little transparency regarding the inner workings. Therefore, their application can be expensive, unpredictable and generally unreliable for a number of use cases (e.g. in live broadcast). To address this aspect, the talk will also illustrate how optimisation of learned models can be used in video processing, so that systems that incorporate these models can be used in a more trustworthy manner. Specifically, video compression examples will be presented, showing how bitrate savings and reduced complexity can be achieved by simplifying relevant learned models.
Biography
Dr Marta Mrak is a Lead R&D Engineer at BBC R&D and an Honorary Professor at Queen Mary University of London (QMUL), Multimedia and Vision Research Group. Within the IEEE, she is the Chair of SPS’s Multimedia Signal Processing Technical Committee, and a member of CAS’s Multimedia Systems and Applications Technical Committee. Her research is focused on video compression fundamentals, new content experiences and data analytics. Marta served as General Co-Chair of IEEE ICME 2020 and was the Lead TPC Chair for IEEE ICME 2019. She has organised numerous workshops and special sessions at conferences, in addition to special issues in journals. During 2013-2018, Marta was an Area Editor for Signal Processing Image Communication journal. She was also appointed as Associate Editor for IEEE Trans. on Multimedia in 2018, and an Associate Editor for IEEE Trans. on Image Processing in 2019. For more details see https://www.bbc.co.uk/rd/people/marta-mrak.