An Interview with Jonatan Samuelsson of Divideon, Winner of IET’s Best Young Professional Award 2018
Tell us a little about yourself and what you like to do outside of work.
I'm a computer scientist who started to work with video processing and compression a bit more than ten years ago. It was in the visual technology group of Ericsson Research that I first came to study the advanced and interesting algorithms that exist inside a video codec and it was with the same group that I contributed to the standardization of High Efficiency Video Coding (HEVC). Outside of work I enjoy spending time with my family and I'm very keen on planning activities we can do together such as going swimming in the summer or going skiing in the winter.
Are you an avid consumer of visual media yourself? If so, are you frustrated by compression artefacts?
Yes, I'm a big fan of streaming services where I have access to a large catalogue of movies and series. It is not uncommon that my wife and I binge-watch an entire season of a series when we find something we both like, and yes, I sometimes get frustrated by compression artefacts. But I don't just get frustrated, I start thinking about what caused the artefact and how it could have been avoided with different encoder algorithms and compression tools.
You have spent most of your career in video compression, what do you find so appealing about this?
When I first came across the problem of video compression I was intrigued by the elegance in some existing compression methods but expected that little more could be done in the area. But the more I've studied it, the more I've discovered that there are so many factors that affect the trade-off between compression efficiency, computational complexity and visual quality. And when I think about the massive amount of video data that is distributed and consumed every day it is exciting to estimate the impact every little performance improvement can have on a global scale.
You have been heavily involved in international standards, what aspect of this do you find most challenging?
I have very positive experiences from working within standardization. I especially appreciate the video-codec-developing organizations MPEG and VCEG and their joint projects, that have produced AVC, HEVC and are now working on the VVC standard. These groups are very technically focused, well structured and very collaborative. Different companies and organizations look into every detail of the video codec and make sure it is high-performing and well suited for all targeted applications. It is a challenge to get a proposal adopted because it has to withstand the thorough review and validation experiments performed by the group.
In 2017 you co-founded Divideon, what motivated you to do this?
I had thought about the idea of starting up a company from time to time but had previously not found the right opportunity. In 2016 it was becoming apparent that the video codec landscape was changing and that many companies were experiencing problematic licensing around HEVC and holding back the deployment of the codec. One of my colleagues at Ericsson and I started to formulate a different approach to tackle these problems and decided to launch Divideon with the vision of developing a high-performing video codec that would not get stuck in the same deadlock as HEVC.
In retrospect, what key advice would you give to someone wanting to launch a start-up?
If you want to create a technology start-up, I think it's important to have a strong vision of what you want to accomplish. Preferably it is something that differs from existing products and challenges conventional assumptions. But perhaps even more important is to ensure that the founders are aligned on where you would like to go, and how to best spend the resource to get there. In a small company it's useful to be able to be flexible and focus your attention on what is most important for the moment, even when it's outside of your typically area of work.
Codec licensing has influenced the design of your xvc codec, do you think that technology or commercial negotiation is the better way for resolving licensing disputes?
In the xvc codec, we have put in place a framework for disabling coding tools without the need for interaction between coder and decoder and without the need for updating the decoder. With this framework, it is possible to stop using a coding tool if it is discovered that it is covered by a patent from a third party that is unwilling to provide a licence for it. We do not consider this framework to be a replacement for negotiation but rather a method to bring more balance to an unbalanced licensing landscape.
Do you think that video compression standards will die out?
No, I think that there are a lot of advantages with the unity and the worldwide reach you get with international standards. I also think the process for producing standards that has been developed and refined in MPEG and VCEG is very efficient and delivers good technical outcomes. That is also one of the reasons why we are contributing to this work and trying to ensure that the next generation codec from these organizations, VVC, is built to support the same kind of framework for the coding tools as in xvc. So far the response has been very positive and although the standardization work of VVC will be continuing for two more years it looks as if this principle will be applied throughout the standardization process.
Have advances in compression become something of a 'Moore's Law' or do you foresee progress running out of steam within a decade or so?
That is a very tricky question but no, I do not foresee advances in compression running out of steam within the next decade. With increased availability of larger screens and with the improved realism provided by higher frame-rates, consumers are constantly expecting higher quality and the only way to meet these expectations, without a substantial increase in delivery costs, is to rely on more efficient compression. Modern video codecs are already complex systems with very many coding tools, but as long as there is a need for better compression there will always be an opportunity to add more coding tools. For example, why not introduce more genre-specific coding tools targeting specific types of content, possibly combined with machine learning to automatically adjust the coding tools to the content?