Connection between Compressibilty % and Average Quant

[Cynthrax]

Hi Guys,

I wonder if it would be possible to define a formula which translates the Percentage of a XviD Compressibilty Test into Average Quantizer of the XviD AVI.

100% Compressibilty would mean average quant 2.0, so perfect quality right?
0% Compressibilty would suggest average quant 31.0, worst quality, but this fact is actually not true. Why?

How to calculate it for other Percentages? I dont know how the average quant changes exactly depending on the compressibility of a file and how the formula behind the Comp. Test of Gordian Knot works?

[Brother John]

Why would you want to bother with quants when you already have a comptest value with a better connection to actual visual quality?

Enough empirical data provided it would of course be possible to estimate the correlation function between quant and comptest percentage. Though I doubt this would be of much practical use.

[Cynthrax]

Why is the Compcheck-Percentage a better indicator for quality? You only know that you reach 60% or 70% Quality of what the XviD Codec is capable of. But with the actual average quant. i know exactly what the quality would be. So I could for example encode my movies only with average quants between 2 and 3,5 and I could always choose the resolution needed for that.

[Brother John]

You’re right, that “better connection to quality” actually was a little misleading. CT percentage and average quant are essentially the same because 100% ct = estimated size at CQ2. So by choosing a percentage you implicitly choose an average quant as well. Which will fluctuate significantly throughout the movie in a 2-pass encoding.

Only CT as an indicator is a lot better supported by software tools and the connection between CT and quant has never been analysed for ASP codecs. From my experience and with the default GKnot style CT I’d suggest you try 75–85% for your quant range.

You’re aware though that frame resolution also has an impact on visual quality, so that lower res, lower quant is not always better than higher res, higher quant?

[Cynthrax]

Yeah, I'm aware that the resolution has an impact on the visual quality. But thats the question before every encode. How low should I keep the resolution to reach my CT-% or should I choose a higher resolution, which then results in lower CT-% and higher quants? So where do I draw the line for quality between Reso/Quants?

[Brother John]

I stopped resizing altogether a long time ago. Before my personal rule used to be: never use a width lower than 600. And indeed the few movies where I broke this rule are the worst encodings I own.

[Cynthrax]

I don't understand your personal rule. I had movies with CTs around 35-40% (target size was 700MB, so 1CD). With a width higher than 600 the final AVI had an average quant of about 6! How can a movie with average quant 6 look good? I did comparisons then between the same avi with 5xx and 6xx resolutions and I could clearly see the artifacts and macroblocks in fast scenes in the 6xx AVI.

[Brother John]

If res is low and quants are still too high, increase the target size. You might have another philosophy, but mine always was: if in doubt, add another CD.