One-pass constant quality mode

[ajmoss]

I understand it is possible to measure the quality of a DivX file recorded in one-pass or two-pass mode. The quality corresponds to the equivalent setting for one-pass constant quality mode, where 100% is two, and 0% is 31. This provides a quantitative way to guage the fidelity of a recording.

Firstly, is it the case that quality (as defined here) corresponds directly to subjective quality? For example, if I adjust the bit rates of all my DivX codings so as to give a constant quality of 60%, is it objectively the case, or is it subjectively likely, that I will perceive all the recordings as being of the same standard?

Secondly, someone published a utility a few months ago that measures the quality of a DivX coding. I can't remember whether it works as a stand-alone application, or with VirtualDub, but it's certainly useful for people who don't use GKnot. Can anyone remember when it is called?

[-h]

All the quality setting changes is the quantizer. The quantizer is the variable that discards information during encoding - the higher the quantizer, the more information gets discarded.

This doesn't however relate to subjective quality. A sequence stored at quant 5 may look just as good as another sequence at quant 3, due to subjective concerns in our perception.

-h

[OvERaCiD23]

http://jonny.leffe.dnsalias.com/

i believe this is the utility you were talking about (only comptest utility i remember at least), but it merely creates a logfile which you load into GKnot for a comptest.

as for the quantizer and quality issue, it can be quite hard to notice a difference in encodings of equal quality (say, comparing Q4 to Q6). the codec is doing the same thing in 1-pass and 2-pass, except for 2-pass it analyzes the video first to distribute the bits. then it encodes it. 2-pass is only useful for obtaining the highest quality for a particular file size. if you don't care about file size (say, storing the videos on your HD), 1-pass is the way to go. also, don't forget, the setup of your frameserver will alter compressibility and quality of videos greatly!

on a side note, why not use GKnot? you don't have to use it to do the actual encoding (even though all it does is send the script to VDub, i still setup VDub manually though), but it's top-notch for a bitrate calc, AR calc, and frameserver creator. plus for the comptest, you can load a first-pass file into GKnot, and it will tell your the average quantizer (good way to judge compressibility). it just makes things as simple as possible, which is always good ;)

[jonny]

Yep CompTest only create the log file for GKnot (it was useful when there was no support for DivX5 in GKnot).

After this i've written DivX5Enc that have is own stand alone comp. test (and yes, work with VDub).


You can't compare the quality value you find in the codec parameter setting with the result of a comp.test.

Yes, there is a relationship between comp. test value & final average quantizer, but is something like this:

http://jonny.leffe.dnsalias.com/g2.gif

Red, green & blue are 3 different clip.
With every clip i've slowly incremented the bitrate and every time i've calculated:
1-comp. test value
2-final average quant encoding in 2-pass mode

For example at 50% the final average quant something near 4.


***EDIT***
Final average quant is calculated summing the quant in analyse.log & dividing by the n of frames (this is the correct final quant. Final quant at the top of analyse.log refer to the first pass and is not accurate. Since 1.07, DivX5Enc report the correct value)

[ajmoss]

[QUOTE]Originally posted by jonny
I've written DivX5Enc that have is own stand alone comp. test (and yes, work with VDub).

Yes, that's the program name I was trying to remember! Thank you.