What's the best lossless YV12 codec currently?

[Stigma]

Sorry for asking, but i wasn't able to find any up-to-date postings on this subject.

I am currently using a really old beta version of VBLE. I think its 0.2 or something. I am reasonably satisfied with it, but im still looking for speed closer to what huffyuv used to give. If there are any newer versions of VBLE, or better codecs that have become availiable recently, i would very much appreciate if someone could point me in the right direction.

(The only other YV12 codec i am aware of is LOCO)

Thanks in advance for taking the time to help =)
-Stigma

[Tommy Carrot]

If efficiency matters the best is ffv1, which can be found inside of ffvfw. It's usually 20-30% better than VBLE, sometimes even more, but requires a very strong CPU for realtime capture. I can encode with 13-15 fps at 768*576 on my athlon 1700+.

If speed is what you care, the latest VBLE codec is 0.4, which is quite close to huffyuv in speed.

[Stigma]

Thanks for the answer. Yes speed is the #1 priority (and allmost the only priority at all).

I think my VBLE version is much older than 0.4, so i lsearch for it in the morning.

Thanks again for the help
-Stigma

[superdump]

Tommy: Please could you inform us of where we can obtain the 0.4 version of VBLE. I'm having little luck after brief searching. :)

[SeeMoreDigital]

I would not mind seeing some kind of 'lossless' audio and video codec list, maybe even with web links!

It would be nice to know where we could read and keep up to date about such ideas.

Cheers

[Sirber]

Hi

I could store them on my server if you give me ZIPs (7z is better :p) :D

[Gaia]

http://www.audiocoding.com/wiki/index.php?page=Lossless+compressed+Audio+Formats

Lossless audio codecs.

[Tommy Carrot]

Originally posted by superdump
Tommy: Please could you inform us of where we can obtain the 0.4 version of VBLE. I'm having little luck after brief searching. :)

This (http://www.cc.jyu.fi/~camneely/vble_video_codec.zip) is what i found, but i'm not sure this is the latest.

[Tommy Carrot]

Oops, the latest version from 0.4 is LOCO, i think there is only 1 version from VBLE. Sorry for the confusion.

[SeeMoreDigital]

>Gaia
Thanks for the link

>Sirber
I dunno. You'll never get to be a Doom9 moderator with that kind of talk!!!! How's your new applicaton comming along?

Cheers

[Stigma]

So, only loco has a new version (0,4) and VBLE is still at the early beta cersion it was first released on?

So do you still recommend VBLE for speed then, or were you also mixed up on this?

-Stigma

[Tommy Carrot]

Well, VBLE is absolutely usable here, i didn't ever have speed problems with it.

[vmesquita]

ffv1 really gives great compression, but unfortunatelly, I can't capture higher than 360x480. In fact this resolution uses about 92%-98% processing time of my athlon 2000+.

[]'s
VMesquita

[Tommy Carrot]

Originally posted by vmesquita
ffv1 really gives great compression, but unfortunatelly, I can't capture higher than 360x480. In fact this resolution uses about 92%-98% processing time of my athlon 2000+.

[]'s
VMesquita

Yes, unfortunately it's not really optimized. It seems to me, the ffmpeg staff is interested in adding support to every codec they can instead of improving the current ones.

This is sad because ffv1 could be a really great capture codec.

[TheXung]

Originally posted by Tommy Carrot
Yes, unfortunately it's not really optimized. It seems to me, the ffmpeg staff is interested in adding support to every codec they can instead of improving the current ones.

This is sad because ffv1 could be a really great capture codec.

I haven't tested ffv1 because I simply don't do TV captures but if someone could give me some filesize test results comparing it to loco, I would really like that.

I have read through the source code of loco and the author describes it as loosely implementing the loco algorithm. However, I cannot find anywhere in his code where he keeps the context counters which leads me to think he has forgotten to implement the context and adaptive correction aspects of loco. Two things that should lead to significant compression gain at very little cpu cost. Reading the code, it seems like loco is just performing median prediction and rice coding the residuals. Now all this discussion leads to this point, a true implementation of the loco algorithm should get you very close to anything out there.

Reading the description of ffv1, I can tell you that it is very similar to loco. The only thing it really has over the loco algorithm is the adaptive arithmetic coder instead of a rice coder. This is like running winrar set to compression best instead of standard - speed and compression wise. However in this case, the compression gain is hardly worth the extra processing.

So if someone can give me a little comparison of the two, I can judge if my hypothesis about the implementation of loco in the codec is correct or not. I have an implementation of the loco algorithm that I've been meaning to throw into a codec but then I saw that someone has already done that. There's no point in me doing it if it is correct, however, I suspect it's not.

[Tommy Carrot]

Originally posted by TheXung
I haven't tested ffv1 because I simply don't do TV captures but if someone could give me some filesize test results comparing it to loco, I would really like that.


The difference varies of course, but ffv1 gives usually 20-30% smaller encoding, with very clean sources, it's even better (for example: loco = 105562k, ffv1 = 62610k).

It's significantly better in any case.

The only problem is: it's too slow to be usable.

[TheXung]

ok, curiousity got the better of me, so I finally found ffv1 and tried it. Didn't quite get 20-30% over loco, more like around 15% on a pretty darn clean source too. AC mode with context 10 only gets ~3% more compression.

However, noticing the speed of ffv1 and loco, I'm also not getting that much of a difference in speed when running in VLC mode. hmmm, I dunno, I think only a significant amount of optimizing ffv1 will make it twice as fast.

[Tommy Carrot]

Originally posted by TheXung
ok, curiousity got the better of me, so I finally found ffv1 and tried it. Didn't quite get 20-30% over loco, more like around 15% on a pretty darn clean source too. AC mode with context 10 only gets ~3% more compression.

However, noticing the speed of ffv1 and loco, I'm also not getting that much of a difference in speed when running in VLC mode. hmmm, I dunno, I think only a significant amount of optimizing ffv1 will make it twice as fast.

Yes, vlc mode is not much slower than loco, but don't forget loco has already half the speed of huffyuv or vble. And ac mode is so slow it's useless.

I think you had bad luck with your test, the gain is usually larger than 15% (and ac mode gives usually 3-10% gain over vlc mode in my experiences).