Mpeg-21 Svc

[Gaborn]

The MPEG seems to be studying a new codec called MPEG-21 SVC.
Supposedly it will be using wavelets for both temporal and spatial redundancy.

According to France Telecom R&D, although it still the very beginning, early tests seem to show a level of performance close to current H264 implementations.

Anyone got some more info about this ?

[unmei]

no i don't know anything, but i found it remarkable that for a search i got top 3 results in korean and fourth and fifth in japanese, then finally at the sixth position one in english.

[Manao]

One of my teachers was also working on a wevelet-based codec to answer the call. Her team claimed, three months ago, to be around 0.5 dB under the PSNR curve of H264. But I wan't able to make her give more informations.

[MfA]

Yeah, everyone is sucking up hard trying to get their codec accepted as the basis for experiments ... microsoft seems to have won though (at least it was going that way a while back, havent payed attention for a while).

If you are interested in this you should subscribe to the SVC email reflector. These codecs are not so much meant to beat h.264, although they mind end up doing so, but to provide scalability (ie. the ability to provide video at scalable rate/quality from 1 bitstream).

To most of us scalability is of 0 importance though.

[Gaborn]

Thx for the answers.

I concur, scalability is not really a hot topic in here, but I wouldn't mind getting scalability as a side effect of using wavelets.
And having a wavelets codec validated by the MPEG could also be a good thing to unify the different implementations of wavelets codecs that seem to blossom lately.

As MfA said, it is primarily aimed at scalability and thus may not be as interesting as I first thought

[unmei]

i don't think it is off topic at all. If you don't get the echo you might have expected it is probably just because people are not exactly excited over the codec. I'm not that excitded for one as i'm interested in either opensource and/or free. Mainly these are small (one-person?) projects. Or if they are h.264 they can also be commercial since there is still a benchmark to be fixed (unlike in ASP where imo Xvid more or less nailed it for some time . So for me, you also won't find me commenting much on real video or wmv..
As for this one i don't know what to think about it since first scalability is not important for my backups and also i thought h.264 were also designed for scalability (another scalable mpeg looks a bit redundant to me).

[Tommy Carrot]

IMO even if this codec doesn't exceed the psnr rate of h.264, it will still have that advantage that it doesn't require additional filtering to hide the artifacts, like the block-based approaches do (and i hate the unnatural smoothness of post- and inloop-filtering). So imo this reason alone makes this research worthy.

[Gaborn]

@unmei: H264 is not designed to be scalable, although it was the case for previous MPEG-4 video codecs.

Damn, I thought that was the place for codec-excited people

[unmei]

hmm, not scalable ..
I thought there were even a special frame type for the transition to a lower bitrate and the possibility to encode in a way that the frame can be decoded with lower quality with only the beginning of a frame's full data. I thought i read this on some commercial encoder's website. But of course these kind of things are not included in any of todays implementations. If you know this is BS please tell me, i might have understood that wrong, or it might have been some vapor-feature (it was a commercial solution's website after all )

[shlezman]

There are many definitions of "Scalable"
Bitrate-Scalability is the possibility to switch to lower/higher bitrate dring transmission. H.264 supports this features using SP/SI (S for scalable) frames, as unmei mentioned.

Temporal-Scalability is the possibility to drop/add frames to change video quality. This is possible with B-frames in MPEG-4/2 and with all frames in H.264

Spatial-Scalability is the possibility to drop/add parts of the frame and change the visual quality. In this type of scalability wavelets have advantage over the "old-fashined" video coding model.

I can only assume that the MPEG21-SVC is aimed to support all kinds of scalability including Spatial-Scalability. I just wonder how ?

[vinouz]

To my opinion scalability is very important, and maybe much more than one thinks at first glance.

What will drive video codec adoption is not only plain DVD or HD DVD use, but also VOD (video on demand) and teleconferencing.
In that area the scalability is a crucial point. And if you have a non scalable codec you have, in multicast services, to transport many versions of the same content at different bitrates.
Which means you're losing A LOT of bandwith.
Considerably more than anything you could have gained by tweaking any codec.
Second, when you have a fine grained scalability, the bandwith you can distribute to anyone is fitted a most to its present conditions, meaning at any point in the time it always has the best possible quality it has.
When the quality degrades or augments smightly, instead of suddenly switching from different streams, with unpleasant steps in visual quality, the video changes little of appearance throughout the variations.
Then, in most situations, the bandwith variations are great and often unpredictable (self-similarity of the bandwith usage prevents it from being easily guessed). So streaming applications always choose a lower bandwith that what is supposed to be left, in order not to be bothered too much by congestion when at little periods of time the ideal bandwith reduces.
Not being constrained as much by bandwith reduction, a scalable codec would allow to permit a higher average bandwith, which is what all of us are pleased with.

Then reencoding such a stream is a joy. Imagine ou have your HDTV and HD-DVD backups on DVDs for maximum quality at home, and you're going on vacations with your laptop/portable video player. Usually you don't have much space left on these device for many films, except if you choose to reencode them at lower bitrates, slightly reduced resolution . This would be doable in a snap with a scalable codec.

Simply imagine : you're having 10Gb to spend on your device. First you put 2 movies. You copy them from your 2 dvd backups downcoding on the fly to fit on this space.
When you will want to add 5 other movies you just have to downcode the two first to gain place, in which you put the 5 remaining.

Isn't it great a tool ?

Isn't that usefulness worth some hypothetic risible compression gains from such a codec ?

I don't think I'm just looking codecs for my little purpose, and I personnaly welcome such new possibilities with warmness and great expectations !

Vincent.