Koepi Xvid 24062003 VHQ Mode?

[Hello]

Could someone please tell me the absolute truth about VHQ Mode? Does every setting (including 0) output the same quality and just differs the compression? And if so, how much time is taken off if you choose the least compression method?

Thanks,

[Hello]

By the way, this is using VDub and VDub MPEG - not sure if that matters or not.

[KpeX]

Could someone please tell me the absolute truth
Hate to burst your bubble, there is no absolute truth, especially when talking about quality.

Does every setting (including 0) output the same quality and just differs the compression?
Compression is not based on VHQ, compression is based on bitrate (desired filesize). As far as quality, that depends on how you quantify quality. Quality is subjective, one attempt to make an objective measure of quality is PSNR (peak signal to noise ratio), tests have shown that VHQ higher settings result in a better PSNR. Use the search button, I believe the results were in one of the Xvid build threads.

And if so, how much time is taken off
Depends on many things. Test it if you want to be sure. Higher VHQ will be much slower, however. Common practice with many people is to always use VHQ-4 unless encoding time is important.

[Hello]

I know compression isn't soley based on VHQ but according to guides it helps compress even more.

[kxy]

Originally posted by KpeX
Hate to burst your bubble, there is no absolute truth, especially when talking about quality.

Compression is not based on VHQ, compression is based on bitrate (desired filesize). As far as quality, that depends on how you quantify quality. Quality is subjective, one attempt to make an objective measure of quality is PSNR (peak signal to noise ratio), tests have shown that VHQ higher settings result in a better PSNR. Use the search button, I believe the results were in one of the Xvid build threads.


I agree with Kpex. If you looked at frequency-banded PSNR, as well
as modifications to PSNR to account for the fact that low contrast scenes will have a much lower MSE for the perceived error
that is because the your eyes and brain are doing contrast adjustments on a region basis. For example, PSNR usually is calculated with a presumed pixel value range of 0-255 [20 * log10(255 / sqrt(mse) )]. What if the image has a range of 50 to 200? Then the formula will change.

[Nic]

They're right, but im sure sysKin could provide a more accurate answer. Basically VHQ is good to use, and use the highest level you can (i.e. because the higher it gets the longer it will take, but also the more cool things like inter4v get used (Does DivX have that yet?))

I must admit I still use VHQ 1, because Im too impatient recently ;)

-Nic

[kxy]

I believe someone made a test indicates that the file size difference between VHQ1 and VHQ4 is less than 1 percent with latest build. And in some cases VHQ4 with qpel and b-frame might cause smearing in the end credit scrolling part(turn the luminance up).

Does this claim still hold?

[Gazza]

Like Nic I tend to use VHQ1. Not just because of speed considerations but, in my mind, using VHQ4 tends to give me a much more flatter colour and contrast than if I used VHQ1. I'll only use VHQ4 if the comp test drives me to do it. In other words, VHQ4 looks washed out to me.

This is only my impression and I haven't really compared these impressions against the views of others. However, I would be interested to see if there are different viewpoints.

[PowerMacG4]

VHQ doesn't affect textures, I thought... only motion vectors...

[sysKin]

Originally posted by Gazza
Not just because of speed considerations but, in my mind, using VHQ4 tends to give me a much more flatter colour and contrast than if I used VHQ1. I'll only use VHQ4 if the comp test drives me to do it. In other words, VHQ4 looks washed out to me. I'm sure you only tried the old VHQ 4 which decreased psnr. The new one has only two effects - one is speed, the other one is making the picture looking as if it had a higher bitrate. That's all.

Originally posted by kxy
I believe someone made a test indicates that the file size difference between VHQ1 and VHQ4 is less than 1 percent with latest build. And in some cases VHQ4 with qpel and b-frame might cause smearing in the end credit scrolling part(turn the luminance up).What do you mean? Filesize is exactly as you put it, unless rate control is b0rked, which is not affected by VHQ anyway.
Af for smearing - I don't think VHQ has anything to do with it. In fact, if you do end credits right, you have a prefectly sharp picture with no smearing (not a bit) at below 150kbps (at 688x resolution).

I do end credits separately and they look better than movie (lol) and take about 5MB, which is just nothing.

Radek

[Nic]

(I think by filesize, he refers to when a piece of video is encoded at a particular quant, and then the filesize indicates how well it compressed)

[temporance]

Originally posted by sysKin
I'm sure you only tried the old VHQ 4 which decreased psnr. The new one has only two effects - one is speed, the other one is making the picture looking as if it had a higher bitrate. That's all.How do you make the picture look as if it has higher bitrate? Does the new one also increase PSNR then, or does this use some weird HVS effect?

[ominte]

How do you make the picture look as if it has higher bitrate? Does the new one also increase PSNR then, or does this use some weird HVS effect?
Syskin was refering to an increase in PSNR if I read his comments correctly.

Higher bitrate = higher PSNR
Higher VHQ level = higher PSNR

[sysKin]

Originally posted by temporance
Does the new one also increase PSNR then, or does this use some weird HVS effect? Yes, almost all it does is PSNR increase. Filesize difference at the same quant is just a side effect without bigger importance - actually the ratio between filesize gain and psnr increase is called 'lambda' and can be set in the sources to anything you want (but the current value looks close to optimal).
Originally posted by Nic
(I think by filesize, he refers to when a piece of video is encoded at a particular quant, and then the filesize indicates how well it compressed)
Yes I know ;) but I'm doing all I can to convince everyone that it's NOT a good way of looking at things. :)


Radek

[Gazza]

sysKin,

I've run a couple of tests and the latest version with VHQ4 has definitely improved on the 'washed out' effects I saw earlier. The first tests where with some dark sources though. I've got a couple of more tests to run (lots of action, bright colours, etc). On what I've seen so far the new VHQ does a nice job (even though it eats up lots of time...).

Cheers

Gazza

[kxy]

Originally posted by sysKin
What do you mean? Filesize is exactly as you put it, unless rate control is b0rked, which is not affected by VHQ anyway.
Af for smearing - I don't think VHQ has anything to do with it. In fact, if you do end credits right, you have a prefectly sharp picture with no smearing (not a bit) at below 150kbps (at 688x resolution).

I do end credits separately and they look better than movie (lol) and take about 5MB, which is just nothing.

Radek

Sorry to mislead you, Radek. I was in a hurry. I think Nic pretty much explains it. What I meant is comparing the 1st pass file size(quant 2 at VHQ1 & VHQ4), the dfference between them is very small.
Are you saying that is not important to compare them? I thought compression test is base on this so using compression test as a guide is not recommanded? :confused:

For smearing, I have download your proper credit encoding clip on irc. I still see the smearing, not a whole lot but it is there. I think the difference is our monitor, my monitor is very bright, and I have my luminace set on a higher range, since I use lumafiler(-#) and unfilter(-#) in my avs to help solving the dark block problem.
Can you please turn your monitor brighter and see if you see what I observed?

I thought Lambda is set somewhere around 1.00, and that only has the effect on constant quant.

Maybe I am just too picky. :D