Keynote Lectures

Seeking Informativeness in Multimedia Data
Amarnath Gupta, University of California San Diego, United States, United States

Why is Zoom So Much More Popular than Standards-based Videoconferencing?
Henning Schulzrinne, Columbia University, United States, United States

From Research to Practice and Back Again: The Role and Benefits of Testbeds for Computer Vision
Noel E. O’Connor, Dublin City University, Ireland, Ireland

Abstract
What makes an image or a video more or less informative? Is there a way to generate an informative image or video? Unlike information theory which is based on the statistical properties of a data item, informativeness is a utilitarian concept. that deeply depends on the contextual information associated with visual objects. This presentation will explore the concept of informativeness for multimedia by adapting an idea from natural language processing. It will make the case that studying information substitutability may give us a clue of evaluating the relative informativeness of multimedia objects, and hence may provide an insight into the generation of synthetic multimedia items.

Abstract
Video conferencing has been around since the 1990s, first based on ISDN for digital phone systems, then H.323, and then SIP for the Internet. The underlying assumption was that phone calls and group communications would all rely on the same standards, creating a universal and seamless user experience.  But even as COVID-19 made Zoom a generic term, these standards-based solutions never got much traction. Even for the popular applications, WebRTC seems to have made modest inroads.  In this talk, I'll try to reflect on the reasons for this development, and why we other standards-based internet applications such as web browsers and email have been more successful. Video conferencing also shares this "fate" with text-based chat, where standards-based solutions are competing with WhatsApp, Slack and other proprietary platforms. I believe that transition from (room) hardware to downloadable software, control protocol complexity, industry structure and communication patterns are at least partial explanations for this development. Finally, I will speculate on how video interaction might evolve in the next few years.

Abstract
Functional testbeds can be hugely important in helping to drive forward progress in certain research disciplines, with topics such as future telecommunications networks and the Internet of Things being notable examples of success in this regard. A testbed's primary role from the perspective of a researcher is typically as a means to make data available to fuel an existing research agenda. However, the opportunity for testbeds to influence research agendas is often overlooked, but when this happens it can lead to a "virtuous cycle" of innovation that can benefit multiple stakeholders. Arguably computer vision has not been well served by testbeds to date, for a variety of practical reasons, not least of which are those that arise due to privacy and GDPR concerns. In this talk I will describe a successful example of how Croke Park, an Irish  sports stadium, and the 3rd largest sports arena in Europe, was used as a computer vision testbed and how this experience positively influenced the research agenda in my group. The experience has led not just to high quality publications but also to successful collaborations in other sectors, including helping our airports deal with Covid-19, and a spin out company, where all work is carried  out in strict adherence to GDPR and privacy concerns.