Gordian Knot Compressibility Test With Divx5/6

[mectron]

Here is my problems:
video file (dvd source) give me as .409 at the compressibility test with Divx 5, but the excat same file give me .491 with Divx 6!!!

the codec were tested using the same file and both used a quantization of 2 (with i think is the standard?)

the result is a estimated file size of 1267 MB for Divx 5 and 1452 mb for Divx 6!!! (this include AC3 audio for both) and each individual test as been done a freshly installed machine

What am i missing? Divx 6 supposed to be better and not produce Bigger file!?

And yes i know that Divx 6 need to be set manualy.

[mimungr]

I noticed the same thing:
http://forum.doom9.org/showthread.php?t=97201

I posted on both the doom9 and divx fora, but received no response. I can't think of any plausible explanation for this.

[Sharktooth]

You're missing the fact that quantization scaling in divx5 and 6 is different.
The same thing happens when you use different quantization matrices on xvid... it's perfectly normal.

[len0x]

yep, that means that with DivX6 you should be aiming at *much* lower compressibility results (<60% actually).

[mimungr]

You're missing the fact that quantization scaling in divx5 and 6 is different.

Yeah, sure, that's possible, but it doesn't seem likely.

One must assume that the quantization in DivX 5 is at least as good as the ITU reference model, which is "nearly optimal subject to the constraints of uniform scalar quantization with a dead zone."

So perhaps the DivX folks really did put a lot of effort into devising an improved non-uniform quantization scheme for DivX 6 (a la XviD's trellis quantization). Does it seem likely this would yield a higher bit rate? I think not. After all, if your sole goal is to maximize quality, H.263 quantization is pretty straightforward. All the innovation centers around reducing bit rate without sacrificing quality.

Furthermore, as I wrote in my earlier post, the SSIM scores declined. Now, you may counter that SSIM doesn't tell the whole story, and we must consider perceived quality instead. But the DivX folks have demonstrated repeatedly, on this forum and others, that they target metrics heavily. Almost all the marketing around DivX 6 has stressed the improved metrics.

In any case, their silence in response to my posts suggests they haven't a good answer. There are DivX employees who are quick to respond to easy questions in this forum.

That said, clearly the DivX folks wouldn't release an encoder that is inferior to its predecessor. They are professional software engineers who know far more about video encoding than I. So I'm confident there is a rational explanation. I just haven't heard it yet.