LCEVC has quickly become the dark horse in the codec race and, according to V-Nova’s Fabio Murra, SVP of Product and Marketing, realistically represents the only new standard deployable this year.
Discussing a paradigm shift in codec standards with the low complexity enhancement video codec (aka MPEG-5 Part 2) as part of the virtual NAB Show Express, Murra described the evolution of the video codec scene in recent years as a “bloodbath”. Such a chaotic scene of disparity, fragmentation and royalty feuds is precisely why we think LCEVC’s name is so apt – both technically and from a marketing standpoint.
“Adopting a multi-codec strategy is not necessarily easy. You need to think about what to deploy, where to deploy, and when to deploy,” outlined Murra. They way he sees it, there are three main decisions to make – regarding compression performance, running costs, and compatibility of the ecosystem. LCEVC promises to address all three, but of course complex questions remain about the low complexity video standard.
Murra pinged off question after question frequently addressed by V-Nova, “How much can a codec improve the quality of my service or running costs? How easy is it to deploy in my infrastructure, or even pubic cloud infrastructure? Do I need to build a whole new data center to deploy this codec? How compatible is this codec with new devices?”. These are not questions unique to LCEVC but are ones its rise to prominence is helping to answer by reducing costs and increasing quality on a simple software-deployable basis.
Roughly, every generation of video codec halves file size (compression efficiency) and increases the time to encode (computational complexity) by between eight and ten times, according to Murra, but typically with the caveat of computational costs rising by an order of magnitude higher.
“This is really the target and rationale behind LCEVC, to impact both of these dimensions,” he said, describing the 2018 proposal from MPEG, which simply (or not so simply at the time) required a codec-agnostic tool capable of enhancing any codec to a compression performance close to its next generation, with same device compatibility without breaking anything. In other words, a codec with low enough complexity to be deployed in software, that can enhance – not replace – existing codecs.
The low complexity part of LCEVC is demonstrated by the use of fewer encoding techniques than other codecs. HEVC, for example, uses over 100 tools and techniques for compressing images, compared to LCEVC which uses less than 10 techniques across the two sublayers – including entropy decoding, including inverse quantization, inverse transform, and smoothing filter.
Murra brought up an image of encoded residual data, a grainy drawing akin to something produced by a 5-year old using an Etch A Sketch. But, of course, this is actually an image containing highly-detailed, sparse information – data which is mostly impossible to infer with an upsampler without the original information. LCEVC therefore uses different tools and different transfers transmitted across the base stream to reconstruct the image. He flipped to an image with lossless 540p upsampled with Lanczos, compared to the same 540p image brought up to 1080p with 300 kbps of LCEVC, with the latter showing a clearer image of an advertising billboard on a race track, alongside a clearer image of icons on a GPS device screen.
Results of a recent test show LCEVC x265 winning against native x265 in 99% of the clips on Netflix’s VMAF (Video Multi-method Assessment Fusion) quality metric, and 61% on PSNR (peak signal-to-noise ratio). Here, the combined LCEVC-HEVC stream was lighter to encode and decode than the native format at same resolution.
But what about the use of LEVC in live/linear constant bitrate (CBR)? Live is LCEVC’s native mode, for which V-Nova recommends encoding in CBR, a simpler configuration which means no implementation of non-default settings in the enhancement layer is required. Murra pointed to a graph showing constant troughs in ‘watchability’ (measured by the Multi-Scale Structural Similarity Index) in the native form, compared to LCEVC Enhanced. “Generally, when the quality is better, we can get extra detail and – more importantly when things get really bad – streams are more resilient. A LCEVC stream would not collapse in blocky artifacts, for example those typically seen in block-based codecs,” he added.
As well as results showing impressive improvements in QoS, arguably the most alluring figure is that LCEVC can reduce encoding and CDN costs by up to 50%. There are more than just lab tests though, with early implementations out there in the form of libraries available for Windows, Linux, and ARM-based (Android, iOS) devices, and of course the V-Nova platform itself.
As for the all-important commercial encoder, Murra could only say commercial deployments supporting LCEVC were “coming soon”. What remains uncertain is the dreaded royalty scheme. Huawei, Samsung, and Qualcomm have all thrown weight behind the recently approved essential video coding (EVC) standard as part of MPEG-5 Part 1 and the three tech titans recently reassured that they will offer “fair” patent royalty terms, for which licensing terms need to be published within two years.