In the second of two articles about the IBC’s long-running Technical Papers Programme, David Davies looks at some of the prominent technologies and topics among the 2024 submissions.
In the first of two articles about the IBC’s Technical Papers programme, we looked at the objectives and structure of the annual ‘call for papers’ and ensuing selection process and conference. In this second feature, the focus turns to the ‘class of 2024’, which was eventually filleted down from more than 300 proposals to around 40 whose authors were invited to submit full papers. Of these, a total of 24 will be presented across nine sessions at the IBC 2024 conference.
Read more IBC Technical Papers: Inside the workings of the committee and conference
Once again, as Executive Producer, Technology, Dr. Nicolas Lodge – who works closely on the Technical Papers Committee with its Chair, Dr. Paul Entwistle – notes in his introduction to ‘The Best of IBC 2024’ publication: “The 20 committee members were not only looking for contributions which are highly novel, but which are also topical, analytical, entertaining, educational, well-written, and which have the potential to make a significant impact upon the media industry. Often the best papers will also have demonstrated their success through simulation, prototype development or full practical trials.”
Volumetric video generation
Among the papers highlighted by Lodge in his introduction to The Best of IBC 2024 is ‘Advancements in Radiance Field Techniques for Volumetric Video Generation: A Technical Overview’ by Joshua Maraval, Nicolas Ramin and Lu Zhang, which appears as part of the ‘XR - advancing, capturing, rendering and delivering’ session. The paper seeks to find a solution to the complexity of capturing and rendering volumetric video – an issue that, it is claimed, is limiting the mainstream creation of VR content.
After a clear review of the various approaches employed to generate 3D video of real-world scenes by employing a large collection of 2D video camera views, the team outlines its enthusiasm for a 2020 paper about a new field called Neural Radiance Fields (NeRFs). As well as explaining the operation of this, they expand upon these ideas and undertake a comparison with several of the best methods from current research materials.
Lodge notes: “It was from this comparison that they demonstrated the significant advantage of a new NeRF-based method called Space-Time Gaussian Feature Splatting, which provides excellent quality rendering and proves rugged in the presence of difficult scene geometries.”
Another of the XR session papers highlighted by Lodge was produced by Shanghai Jiao Tong University. “Their approach has been to use pre-trained NeRFs for the real-time representation and rendering of 3D human heads (including facial expressions) at a mere 5kbit/s,” he says.
AI as an enabler
It will come as no surprise at all to discover that AI is very well-represented across the submissions, although in a significantly more varied way than would have been the case pre-pandemic. “We’ve seen AI emerge literally over the last six or seven years from what I would call a curiosity on how to improve metadata, to something that is now in use across the industry,” says Entwistle. “I know at this point in time that generative AI and large language models are highly topical subjects, but there is a lot more to it than those two techniques. And so if you look across our sessions, you’ll find AI in many different sessions including production, video coding, XR, and audio and speech. So in that time-frame, you can really see how the technology has spread throughout the industry and is being used well.”
Quite how far things have progressed can be gleaned from the three-paper session entitled ‘AI in Production – training and targeting’, which will feature presentations from representatives of the EBU, Viaccess-Orca and NHK. The first paper, from the EBU, outlines a facial recognition machine learning system that has been tailored to the specific needs of media documentalists. It seeks only to identify ‘active’ and not incidental characters – assisting in compliance with privacy regulations.
The second paper, from NHK, explores a solution to the over-training problem when applying machine learning to multi-label indexing of news items – a particularly important consideration for infrequently classified events. Then, in the final paper, Viaccess-Orca envisages a future where generative AI is employed to adapt advertising based on the characteristics and beliefs that have been gathered from the target audience.
“AI is what you call a supporting technology – an enabling technology – and that’s what makes it so special, I think,” says Lodge. “I was surprised to look back to 2019 and find that even 50% of the paper submissions then used AI in some way. But now it’s being used in just about everything – even things like image compression that you think are very algorithmic. People are finding all kinds of ways to apply the learning techniques of AI to improve things.”
Media provenance
In a year characterised by conflicts around the world, and with important elections set to take in place many countries, the ability for viewers and readers to trust the news they access is of paramount importance. Both Entwistle and Lodge acknowledge the timeliness of the ‘Provenance – what can we trust?’ session, which comprises two papers exploring methods of building trust and ensuring the authenticity of content.
“Developing mechanisms through which we are able to trust distributed media, especially news, is one of the most fundamental issues in our industry and in democratic society, more widely,” says Lodge. “We present two papers on this topic, one by the BBC which looks at standards for content credentials and which reports on a trial assessing subsequent gain in audience trust. The second, by Verance Corp in the USA, provides a more technical approach where watermarking and cryptography are discussed in some depth. Both papers relate to standards emerging under the Coalition for Content Provenance and Authenticity (C2PA).”
Read more IBC2024 Accelerator Project: Design Your Weapons in the Fight Against Disinformation
Having previously lamented the lack of strong proposals about sustainability, Entwistle is “really pleased” by the strength of the papers in this year’s session, which include a presentation from Arqiva and the BBC about an approach to reducing the energy required to transmit terrestrial DAB. He also draws attention to the two-paper streaming session, where streaming is examined from ‘both ends’.
In the one paper, reveals Entwistle, AWS “looks at the technical challenges of mega-concurrency in video streaming, driven by cricket in India, where the sport drives huge traffic surges (10 million connections in 10 minutes) and a record 59 million concurrent connections is held. We all take for granted the fact that we can stream videos on headphones and televisions and across the internet. But from an engineering and technical point of view, how do you do that at such a massive scale? These are really large technical challenges that companies have to cope with on the back of streaming.”
At the ‘other end’, the BBC will discuss its study of actual day-to-day user experience of mobile services while on the move and using its BBC Sounds app. Correlating network/radio properties and CDN logs, the work aims to determine the concept of a ‘service coverage’ area.
A fuller flavour
Invited for his primary take-away from this year’s papers, Entwistle says that “the main thing is the breadth of technology, and it’s really impressive. Our industry covers so much now – from 5G and how it’s impacting production and delivery, to new content formats from extended reality, and so on. The challenge with breadth is the interoperability between all those things to make a system work, but as a researcher or innovator, you are probably working within a fairly small field. However, when you come to the Technical Papers Conference, you can gain a flavour of all the other parts of the technology fields within our industry.”
For his part, Lodge indicates that – with creative uses of AI being so prevalent now – the papers may also reflect an industry in the midst of its most profound period of change for 30 years. “When you look at the influence of AI, this all-embracing technology which applies to so many fields, it’s much like the change from analogue to digital in terms of its impact. AI could be the biggest advance since that transformation, and in terms of its ability to influence the work and the workplace, it could be even bigger. So it’s a very exciting time right now and I’m very pleased that the papers we shall be hearing at the conference reflect that.”
For a full list of topics and the day-to-day schedule for the Technical Paper Conference programme taking place in Conference Hall 2 at IBC2024, please click here.
No comments yet