5.2 and comp test

[len0x]

Is it just me or 5.2 is totally buggered with respect to single pass quality-based encoding which is used for comp test?

I started with no options and profiles selected and MPEG matrix. Previously we had to use quant=2 for comp test. So I ran my tests and every time I got more than 100%! It looked like it doubled comparing to 5.1.1. OK, i said lets try quant=1 (may be they changed it). Then I got sort of normal values like in 5.1.1.

Then I switched on adaptive single bframes. And I got less compresibility than without them! (by 3-5%) I couldn't believe my eyes...

Anyway then I decided to switch to H.263 matrix. Comp test percentage dropped dramatically almost by 50%! So I switched back to quant=2 and managed to get values closer to what 5.1.1 was giving with b-frames (and h.263) previously.

So, does anyone else have these issues? (i.e.:
- single bframes producing larger size output
- mpeg and h263 matrixes using completely different quantizer settings )

P.S. problem seems to be only in quality-based single pass encoding as two passes with the same options result in good encodes (no undersize in case of comp test value > 100%)

*Edit* it seems that if H263 is selected everything is as it used to be in 5.1.1, but with MPEG matrix everything goes out of control...

[Sharktooth]

5.2 MPEG quantization is quite... b0rked.. i hope they fix it soon.

P.S.: WB Lenox :)

[Gannjunior]

Originally posted by len0x
......P.S. problem seems to be only in quality-based single pass encoding as two passes with the same options result in good encodes (no undersize in case of comp test value > 100%).....

Hi Len,
in two passes, MPEG2 Quant hasn't any problems, 'cause codec gets Q=1, so you can reach your desired size, only until you hit mpeg2's Q=1 saturation... and this is not too much difficult ( mpeg2 matrix size at Q=1 arrives nearly h263 matrix size at Q=2...).in the pratice, mpeg2 is competitive only at very low bitrate.but, if you do a little test at medium-high bitrate, at parity size, compared with h263, mpeg2 matrix loses the comparison whether by "eyes" or SSIM values...i've verified it by a long session of tests ( that i'm going to complete).
So, as for you and Shark, I'm conviced too, mpeg2 matrix has something wrong...
:confused:

ciao!

[Gannjunior]

Problem not solved in 5.2.1 too....I reported the problem in divx.com forum too some weeks ago, but with any answer... ->HERE (http://forums.divx.com/viewtopic.php?forum=5&topic=60531)

In this way, i don't see any reason to not to use permanently VP6 in my encoding... ;)

ciao!!

[Taurus]

Divx 5.2.1:
Encode Performance: Standard, No Options
1-pass quality-based
Quant = 2
Bidirectional encoding = off
H.263 = 47.5% Compressibility test
---------------------------------------------
Encode Performance: Standard, No Options
1-pass quality-based
Quant = 2
Bidirectional encoding = off
mpeg-2 = 75.6% Compressibility test
---------------------------------------------
Encode Performance: Standard, No Options
1-pass quality-based
Quant = 1
Bidirectional encoding = off
H.263 = 47.5% Compressibility test
---------------------------------------------
Encode Performance: Standard, No Options
1-pass quality-based
Quant = 1
Bidirectional encoding = off
mpeg-2 = 75.6% Compressibility test
---------------------------------------------
Encode Performance: Standard, No Options
1-pass quality-based
Quant = 2
Bidirectional encoding = 0n (adaptive single consecutive)
H.263 = 53.3% Compressibility test
---------------------------------------------
Encode Performance: Standard, No Options
1-pass quality-based
Quant = 2
Bidirectional encoding = 0n (adaptive single consecutive)
mpeg-2 = 82.0% Compressibility test
---------------------------------------------
Encode Performance: Standard, No Options
1-pass quality-based
Quant = 1
Bidirectional encoding = 0n (adaptive single consecutive)
H.263 = 53.3% Compressibility test
---------------------------------------------
Encode Performance: Standard, No Options
1-pass quality-based
Quant = 1
Bidirectional encoding = 0n (adaptive single consecutive)
mpeg-2 = 82.0% Compressibility test
---------------------------------------------
Just numbers to see the difference.
It's obvious: Quant = 2 is used on 1-pass quality-based, no matter where you drag the slider or what you fill in the textbox.
But strange: compressibility ranges from 47.5% to 82.0%
So my opinion too, keep your hands of from mpeg-2 matrix.
There is something going on strange under the hood :D

Taurus

[Fr4nz]

No news about this??

[LordRPI]

I'm passing it on.