[XviD-devel] bframes 'user' tests

[Christoph Lampert]

Hi,

B-frames direct mode is still too buggy to test quality. 
You can remove the comment from the the line 

//   d_sad16 = 65535;

in src/xvidcore/motion/motion_est.c 

Then direct mode will not be used and you can test b-frames
more reliable.

gruel

--------------------------------


On Sat, 20 Jul 2002, Edouard Gomez wrote:

> Hello,
> 
> I know bframes  are still beta. But as the  bframes support is getting
> better each day, i thought some  real test cases could show "big" bugs
> in the bframe's code.
> 
> Here are the options i've used :
> 
> [export_xvidcvs.so]     multi-pass session: 0
> [export_xvidcvs.so]                quality: 0
> [export_xvidcvs.so]      bitrate [kBits/s]: 710
> [export_xvidcvs.so]  max keyframe interval: 250
> [export_xvidcvs.so]    max bframe sequence: 1
> [export_xvidcvs.so]     bframe quant ratio: 140
> [export_xvidcvs.so]           motion flags: PMV_HALFPELREFINE16
>                                             PMV_EXTSEARCH16
>                                             PMV_EARLYSTOP16
>                                             PMV_USESQUARES16
>                                             PMV_HALFPELREFINE8
>                                             PMV_EARLYSTOP8
> [export_xvidcvs.so]           global flags: XVID_GLOBAL_DX50BVOP
> [export_xvidcvs.so]          general flags: XVID_H263QUANT
>                                             XVID_HALFPEL
>                                             XVID_ADAPTIVEQUANT
>                                             XVID_INTER4V
> [export_xvidcvs.so]              cpu flags: XVID_CPU_MMX
>                                             XVID_CPU_MMXEXT
>                                             XVID_CPU_SSE
>                                             XVID_CPU_TSC
> [export_xvidcvs.so]             frame rate: 25.00
> [export_xvidcvs.so]            color space: YV12
> 
> Test clip     : "The Fith Element" DVD source / frames 0-5000
>                  Original frame size : 720x578
>                  Encoded  frame size : 608x256
> 
> Software used : transcode 0.6.0 pre3 + bframe patch for xvidcvs module
> 
> 1 - BFrame encoding speed.
> 
> Nothing to say here, the loss is just 1 frame/s encoding. Great work !
> 
> 2 - BFrame visual quality
> 
> Well, i know there's still lot of work in this area, i would just give
> my opinion. When the max_bframes >=2 the quality gets worse very quick
> when there's  a bit  of motion. Moreover  the current default  for the
> bquant_ratio seems to  be too high. With a ratio  of 200 (default) the
> quantizer become too big (quality is poor - normal). The 125-140 range
> gives good  results and the bitstream  size is still reduced  by a 1/4
> ratio (IP  encoding is  20Mo, IPB  encoding is ~15Mo  with a  ratio of
> 140).
> 
> 3 - Visual artifacts surely due to bugs
> 
> If someone asks me, i'll post the clip.
> 
> There's a  bug in the code  because some MBlocks diplay  just as black
> squares  even   in  bright  zones  (chroma  is   ~0  too).   Moreover,
> mplayer+libavcodec reports some "Bad MPEG4 header - broken or missing"
> when max_bframes  is >= 2. I  can't reproduce this bug  each time, but
> there's for sure something wrong with the generated bitstream.
> 
> In order to sumarize :
>  - MBlocks are wrong  sometimes
>  - bitstream is not well formed when max_bframes >= 2
> 
> The bitstream error might cause the first one ?
> 
> 
> I hope this could help bframes developers (well i know this bug report
> sux a lot, it lacks of details :-)
> 
> -- 
> Edouard Gomez
> _______________________________________________
> XviD-devel mailing list
> XviD-devel@xvid.org
> http://list.xvid.org/mailman/listinfo/xvid-devel
> 

-- 
Christoph H. Lampert chl@math.uni-bonn.de | Diese Signature wurde maschi-     
Beringstr. 6, Raum 14 Tel. (0228) 73-2948 | nell erstellt und bedarf
Sprechstunden: keine, aber meistens da    | keiner Unterschrift. AZ 27B-6