xvid 1.0.3 vs xvid 1.1 beta1

[Hironimo]

Hi there

I did some comparison between xvid 1.0.3 and xvid 1.1 beta1 in terms of quality. For this I encoded the exact same source with the same settings, except that I added VHQ vor B-Frames in xvid 1.1.
The video wasa 3884 frames from Enterprise 3x16, 720x406@1840 kbit/s 2pass. The source was a decent quality analog capture with some noise-reduction and sharpening filters.

Here the results from the avisynth comparison filter:

xvid 1.0.3:
Filesize: 35140 kb
Frame: 3884 ( min / avg / max )
Mean Abs Dev: 1.8890 ( 1.617 / 3.667 / 2.829 )
Mean Dev: +1.2989 ( +0.625 / +1.339 / +2.829 )
Max Pos Dev: 44
Max Neg Dev: -39
PSNR: 40.07 dB ( 28.56 / 34.46 / 41.34 )
Overall PSNR: 34.10 dB

xvid 1.1 beta1
Filesize: 33750 kb
Frame: 3884 ( min / avg / max )
Mean Abs Dev: 1.8890 ( 1.674 / 3.691 / 2.461 )
Mean Dev: +1.2989 ( +0.677 / +1.366 / +2.461 )
Max Pos Dev: 41
Max Neg Dev: -36
PSNR: 41.05 dB ( 28.62 / 34.40 / 41.05 )
Overall PSNR: 34.06 dB

Interesting to see: 1.1 beta produced a noticably smaller file,
but the quality is almost as good.
When doing the comparisons xvid 1.0.3 was used, if that matters (does it?)

Can you tell me why at the exact same settings the resulting size is smaller for 1.1 beta?

What do you think of the results?

I think i'll change to 1.1 beta1 as it produces better quality at the same sizes I presume, and with about 12% better speed (I did not run exact benchmarks on the speed, but the rough tests confirmed the values).

Maybe others could post thir results and possible bugs or errors in encoding so that others can better choose between the two before 1.1 final is out.

Cheers, Hironimo

[Heini011]

Hi Hironimo,

i wouldn't expect much differences when using a source with decent quality. try a high-quality source and compare with your human visual system too.

>Interesting to see: 1.1 beta produced a noticably smaller file

1. the default overflow treatment (5,5,5) is optimized for a whole film (> 1 hour). use a longer clip for testing 2 pass encoding and use (10,10,10) for short clips.

2. check the stat-files for the first-pass sizes. did you use the default q 1-31 settings or 2-31 ?

i use always 2-31 and set the aimed second pass size 5-25% below to first pass size (from stat file).

>with about 12% better speed

decent quality sources requires strong filtering. xvid is much faster than good filters.

greetings

[Teegedeck]

Compressiblity (filesizes) has to be measured at a constant quantizer, 2-pass is not applicable for that. For such small files, I think the aberration is within the margin of error and probably not determining a general difference.