Help to development of videocodecs.

[IgorC]

Maybe it possible to provide some helpfull information to devs by installing special programm that will use CPU free cycles and will send valueable information through net.

Like those :
http://www.mersenne.org/
http://fightaidsathome.scripps.edu/index.html

[bratao]

Sorry, but how find aids cure help videocodec ? maybe discoverying aids cure, the developers would think less about protection and have more free time to develop
=)

[DeathTheSheep]

No... I think the "aids" is in his sig, and has nothing to do with his actual message ;)

[shevegen]

Its his sig, and it shall rightfully distract people!


But in fact he also made a link, so I must ask myself why
you are connecting something (biology/medicine) that has nothing
to do with Video stuff?
Yes we have SETA and grid computing but how in reality do devs develop?
In rather small groups with solid C knowledge. And hacking hours and
hours.
Plus you were a bit vague.
"will send valueable information through net."
What kind of information would be needed which cant be sent i.e. with a bugreport, or a timed test of something?

[IgorC]

The development of videocodec requeries a lot of stastistic information. It will help to undertand or to see better what happens for example with ratetol, differents combinations of coeficients, boosts. It might be helpfull to optimize for speed i.e. trellis or motion vectors search etc. There are a millions of combinations.

But it's hard to implement and of course harder to get it work in right direction and w/o bugs.

Yes, it was a little bit "crazy" idea. But many companies of auto, soft/hard and another industries implemented this system to predict the better way of evolution.

[berrinam]

I think it's a good idea, but it would need a lot of work to get running. Creating some program which runs in the background and does tests on quality/speed tradeoffs and transmits the results to some centralized server would mean a huge scope for testing, and comprehensive testing of every combination of settings becomes much more of a possibility. It would obviously need a lot of work, though, in developing such a program/interface, but it could really help in finding peak settings (I've done some tests of a similar sort on a much more modest scale). The good thing about such a system is that it runs all the time, and without human input, so it doesn't interfere with the user of the computer in any way.

I think that it wouldn't need to be done by the developers, though. Someone else with coding experience could perhaps design these programs, which would simply run whatever tests necessary on the client-side. The biggest problem, as I see it, is that in the end, encoding is a visual thing, and so optimization of the encoding process cannot be done without some very reliable quality metric -- which we don't have.

[IgorC]

Yes. Metric is another problem. OPSNR has a significant error +/- 0.35-0.5 dB (i.e . 42 and 42.5) . SSIM is better in this term . For my eyes only +/- 0.2 SSIM ( 80 and 80.2 ).
But finally it would be up to devs how to interpret the results.