LCEVC - Low Complexity Enhancement Video Coding

[ksec]

Since there is no thread that is dedicated to LCEVC[0] I may as well start one.

Because I just discovered SBTVD 3.0 [1], the Television Broadcasting standards used in South America has adopted LCEVC usage. That is a whole ecosystem of mandatory support for LCEVC.

Not sure what the baseline Codec yet. May be VVC.



[0] https://www.v-nova.com/lcevc-enhanced-video/

[1] https://www.v-nova.com/press-releases/brazilian-sbtvd-forum-selects-mpeg-5-lcevc-for-the-video-enhancement-codec-layer-of-its-next-generation-television-broadcast-system-tv-3-0-2/

[Selur]

Looks like some glsl sharpening filters,...

[filler56789]

🧐 https://en.wikipedia.org/wiki/LCEVC

It is already possible to encode video files with LCEVC using FFmpeg 4.0.1-17...

😳

[Selur]

Only if you have a ffmpeg version which is build with that encoder,... (haven't seen one around)

[lvqcl]

https://en.wikipedia.org/wiki/LCEVC

Quote from this wiki page:

Keywords for further studding (may require searches outside of wikipedia)

LOL.

[ksec]

Oh wow from Wiki

It will be used to enhance a video base layer encoded with VVC for broadcast and broadband delivery.

So this is VVC + LCEVC!! Oh I want to play with it.

[Jamaika]

Does anyone know where to download opensource lcevc from?
http://mpegx.int-evry.fr/software/MPEG/LCEVC/LTM.git
I found a software-based decoder on github.
https://github.com/eocene317/LTM-master

[Blue_MiSfit]

LC-EVC is definitely neat :)

Think of it like HE-AAC for video, except instead of being locked to using AAC as the base codec (or rather AVC as most commonly used with LC-EVC) you can use whatever base codec you want.

You essentially subsample the video (typically, though you don't have to) and then store high frequency information separately to aid the reconstruction of a higher quality image in the decoder. This way you can spend all your bits to encode a very stable base layer, even if it's soft.

You get added benefits here like often only needing to encode and decode 1/4 to 1/2 the pixels in the base layer, which has clear performance benefits. You can also reduce the bitrate a fair bit, which leads to overall bit savings for a given encode / ABR ladder.

It's not perfect, but there's definitely something to this approach!

[benwaggoner]

And not it's not "just" blind postprocessing like the old WMV and MPEG-4 ASP players. Metadata is generated while encoding that the LC-EVC decoding filter uses to more accurately reconstruct the look and feel of removed detail. And it is set up so that the preprocessing layer can be generated outside of DRM protected memory, and then the resulting bitmap merged with the decoded frames.

[birdie]

Some paper https://arxiv.org/abs/2204.05580

[ksec]

Some paper https://arxiv.org/abs/2204.05580

No wonder why it is making some noise in broadcast industry. It seems LCEVC would do exceptionally well with low movement materials, I am thinking News and possibly anime content. Where it gets 50 to 60% bitrate reduction regardless x264 or x265.

[rwill]

No wonder why it is making some noise in broadcast industry. It seems LCEVC would do exceptionally well with low movement materials, I am thinking News and possibly anime content. Where it gets 50 to 60% bitrate reduction regardless x264 or x265.

Yeah running x264 with the medium preset and x265 with the veryfast preset somewhat matches the quality of broadcast encoders I guess...

The github with the data set is also very interesting, just a license and a readme. The readme states to send an email to maybe get access to the data set...

[Forteen88]

Someone else did a test on LCEVC here,
https://www.youtube.com/watch?v=Frrw8O2ax6I

LCEVC got really good results there too.

[benwaggoner]

LCEVC does have some nice attributes, and does improve the psychovisual appearance of detail. I'm kind of reminded of VP6's postprocessing of years gone by. The core VP6 codec wasn't great by any means, and way too PSNR-is-law. But the decoder's postprocessing did a great job of synthesizing noise, sharpening, deblocking, and deringing. None of that actually adds new discernible vision information, but but did keep the video from looking nearly as "web video" at lower bitrates compared to VC-1 and that era's H.264 encoders.

LCEVC is a more advanced version of that, with metadata that can actually improve real detail, instead of just mask the lack of it. And it is backwards compatible. The base stream remains decodable even if LCEVC isn't enabled on that device. At low bitrates, the metadata can make up a good fraction of the total bitrate, so it would be generally better to stream the base layer independently, and only the LCEVC on a player identified to support it. The overhead is much less material >1 Mbps.

[Jamaika]

Since there is no thread that is dedicated to LCEVC[0] I may as well start one.

Because I just discovered SBTVD 3.0 [1], the Television Broadcasting standards used in South America has adopted LCEVC usage. That is a whole ecosystem of mandatory support for LCEVC.

Not sure what the baseline Codec yet. May be VVC.



[0] https://www.v-nova.com/lcevc-enhanced-video/

[1] https://www.v-nova.com/press-releases/brazilian-sbtvd-forum-selects-mpeg-5-lcevc-for-the-video-enhancement-codec-layer-of-its-next-generation-television-broadcast-system-tv-3-0-2/

I'm looking at an ad for lcevc v-nova. I believe she is outdated.
I don't know how evc differs from lcevc. The evc implantation on their sites is advertised as lcevc. In addition the codec libraries are old from two years ago.

https://gitlab.com/v-nova-public/ETM/-/tree/vnova/ltm
https://gitlab.com/v-nova-public/VTM/-/tree/vnova/ltm
https://gitlab.com/v-nova-public/HM/-/tree/vnova/ltm
https://gitlab.com/v-nova-public/JM/-/tree/vnova/ltm

The same is true for the LTM converter (EVC, VVC, H265, H264). In addition converter is incomplete and has bugs.

How does this relate to the new free codecs on git?

https://github.com/mpeg5/xeve
https://github.com/mpeg5/xevd
https://github.com/fraunhoferhhi/vvenc
https://github.com/fraunhoferhhi/vvdec
https://vcgit.hhi.fraunhofer.de/jvet/VVCSoftware_VTM
https://vcgit.hhi.fraunhofer.de/jvet/HM
https://vcgit.hhi.fraunhofer.de/jvet/JM

[rwill]

The v-nova repo things are just slightly modified reference softwares. Strange, I thought at least ETM availability is being watched over by MPEG.

The free codecs on git are the respective reference softwares except for xeve/xevd and vvenc/vvdec which try to be independent implementations of their video coding standard.

EVC is not really related to LC-EVC by the way.

LC-EVC can be put on top of more or less any video codec, I guess thats why v-nova has forks of the AVC, HEVC, VVC and EVC reference softwares in their git.

[ivanb]

Does anyone happen to have sample files with LCEVC muxed into MPEG-TS or MOV? I was planning to add support for this format as part of this issue (https://github.com/vtclab/analyzer-public/issues/16), but I just can’t seem to find suitable samples around

[Z2697]

Does anyone happen to have sample files with LCEVC muxed into MPEG-TS or MOV? I was planning to add support for this format as part of this issue (https://github.com/vtclab/analyzer-public/issues/16), but I just can’t seem to find suitable samples around

There're some examples in v-nova's website, I managed to download them but I'm not sure if they are "complete" lc-evc streams, probably yes, looks like there's something in ITU-T T.35 SEI

[ivanb]

There're some examples in v-nova's website

That's something I can start with, thanks!

looks like there's something in ITU-T T.35 SEI

The odd thing is that I can’t find any specification that defines this method of storage. I found a description of how LC-EVC video should be embedded in an MP4 container, but it seems like V-Nova is using a different approach.