THISMONTH’S FEED ROUND TABLE GUESTS ARE:
FEED: WHAT ARE SOME OF THE BEST MACHINE-LEARNING APPLICATIONS YOU’VE SEEN LATELY?
BEN DAVENPORT, PORTFOLIO MANAGER, ARVATO SYSTEMS Ben Davenport is a portfolio manager at Arvato Systems, where he supports companies in their digital transformation, particularly with the power of artificial intelligence.
BEN DAVENPORT: The most visually impressive application I’ve seen is an implementation we’re working on together with Nablet for automated highlight package creation for football matches. The software automatically detects match highlights, including goals, tackles and referee interventions, and creates either a final file or a project within our browser-based editor or production asset management system. Another exciting development is around dynamic pricing for advertising inventory. Here, algorithms detect where sales patterns are differing from the norm or a predefined ‘envelope,’ giving sales teams guidance as to whether they should raise or lower package or spot pricing, ensuring that broadcasters neither undersell nor under-fill their inventory. TONY JONES: As UHD TVs become increasingly common, there is a gap in the market for UHD content in order to create UHD channels. We have been applying machine learning to upconvert library HD content to content suitable for UHD viewing, to provide an experience more like native UHD than can be achieved with conventional upconversion methods, such as bicubic interpolation. Using a hybrid GAN (generative adversarial network), we have been able to
produce convincing UHD results from HD content, where plausible new detail has been synthesised to create content that has features similar to native UHD. This can work with both progressive and interlaced HD content and it has the potential to be done in real time, either on-premises or in a public cloud. ARASH PENDARI: The recently released Nvidia machine-learning application is very interesting, I must say, but the one used in DLSS 2.0 can produce some of the most impressive and incredible results I’ve seen out in live production. It’s a technology that takes low-resolution images and converts them into a higher resolution by guessing what pixels should be added where. This can be used to, for example, create 4K images, videos and games. The potential of this technology to lower the bandwidth of 4K video streams or games is extremely valuable. This is also a technology you can use to restore old movies and give them a 4K resolution viewing experience. In the future, I believe machine learning will be able to do this fully automatically. Imagine how, in the future, we could be watching Charlie Chaplin movies with an 8K resolution in colour and 5.1 sound, all thanks to machine learning!
JON FINEGOLD, CMO, SIGNIANT Jon Finegold has been working in the digital content field for 20 years and is now chief marketing officer at file transfer technology company Signiant, working to help customers understand its intelligent file transfer and cloud-native SaaS products.
TONY JONES, PRINCIPAL TECHNOLOGIST, MEDIAKIND As principal technologist for
MediaKind, Tony Jones plays a key role defining core technology for MediaKind, from video processing to TV platforms. He has most recently been developing strategies for high dynamic range, cloud technology in the media space and end-to-end storage and ABR delivery optimisation.
ARASH PENDARI, CREATIVE DIRECTOR AND PRODUCT EVANGELIST, VIONLABS With a background in game
development and game design, Arash Pendari founded Vionlabs in 2010 with the mission to enhance the user experience for TV and movie lovers. Over the last decade, the company has developed content discovery tools and experiences powered by AI.
feedzinesocial feedmagazine.tv
Powered by FlippingBook