VVC Should Help with 8K Distribution but Benchmarks not Clear Yet
Virginie Drugeon, a Senior Engineer Digital TV Standardization at Panasonic, gave a presentation in a webinar to discuss the newest video codec – VVC or Versatile Video Coding. The presentation focused on the key features and tools of the codec and how it is different from HEVC. In a follow-up discussion, we also asked about the implications for 8K.
VVC, also known as MPEG-I Part 3 / H.266, was developed by the Joint Video Expert Team (JVET) of ISO/IEC MPEG and ITU-T VCEG. It was approved in July 2020. Two licensing pools have been set up with the hope of avoiding the complex and difficult licensing options that have prevented wider adoption of HEVC.
Like HEVC, VVC is a block-based video codec meaning it divides the video frame into a series of blocks for evaluation and coding. In many ways, it is very similar to HEVC but adds more tools and flexibility. For example, VVC expands the coding block sizes to 128×128 pixels and offers additional blocking geometries. There are also special tools to support 360-degree video formats and improvements in the many other tools in HEVC. The complex structure and high-level summary of the toolsets are shown in the diagram.
The stated goal of VVC is to achieve a 50% reduction in bit rate compared to HEVC (all parameters being equal). Since encoding parameters and implementations can vary widely, the standards groups use reference software and content to benchmark performance. The reference software for VVC is called VTM while the reference software for HEVC is called HM. The table below shows the results comparing the two reference software encodes using PSNR as an evaluation parameter, which Drugeon acknowledged is not the best way to measure picture quality.
As can be seen, the VTM encoding time (EncT) is significantly longer (on average 8 times longer) than the HM encode time with decoding taking 1.6x more time.
Reference encoders are not optimized for encoding/decoding speed so commercial deployments should achieve much better results. Last October, the JVET group did subjective visual testing using HM, VTM and the first implementation of VVC using the open-source version called VVenC, which is reported to be about 100x faster than VTM. They observed improved opinions of the content with VTM and VVenC vs. HM at the same bit rates.
But this test only used 4K SDR content. What about 8K HDR content we asked Drugeon? She said that no 8K tests have yet been done by JVET but that is something they may want to look into.
VVC also includes 6 profiles which are sub-defined by a tier (2) and level (13) which correspond to different capabilities needed to decode a bitstream. “A level of a tier specifies a set of limits, such as maximum picture size, maximum luma sample rate or maximum size of the coded picture buffer,” explained Drugeon. “Higher levels increase these values compared to lower levels so that decoders that fulfil the limits of one given level can also decode all the bitstreams corresponding to lower levels. To encode 4K video at 60 frames/second, at least level 5.1 is needed, but you need level 5.2 for 4K at 120 frames/second, because the luma sample rate is higher, even if the picture size is the same.” Levels 6.0, 6.1 and 6.2 are needed for 8K content.
Drugeon said that she expects the Main 10 profile, which supports 4:2:0 chroma sub-sampling and up to 10 bits to be the most standard and maybe the most popular, but the Multi-layer Main 10 profile allows scalability meaning you can have a base layer and an enhancement layer combined in a single bitstream. As a result, we asked how she viewed the use of the Multi-layer Main 10 codec for 8K content.
“As for scalability, I would be happy if a good use case is found for it, but historically, it has never been used much, unfortunately. Typically scalability is rather interesting to enable use cases whereby the same bitstream can be fed to very different types of receivers with different capabilities, but it is usually not helpful for data rate reduction compared to a single-layer bitstream at the highest resolution. Of course, there would be a data rate gain compared to simulcasting both the 4K and 8K versions of the same video. That said, your use case of a 4K base layer enhanced to 8K has not been tested so far, and it would be interesting to get test results to see the effect in terms of bitrate.”
As far as commercialization of chipsets to allow commercial deployment of hardware-based VVC solution, Drugeon said this is in development but she could not predict availability.