[MPlayer-users] MPlayer-1.0pre4 bug with divx4 multipass filter

[D Richard Felker III]

On Mon, Sep 13, 2004 at 10:31:28PM +0400, Vladimir Mosgalin wrote:
> On Mon, 13 Sep 2004, D Richard Felker III wrote:
> 
> DRFI>> I've never encountered that bug in the wild, but...
> DRFI>
> DRFI>Because no real release group is so stupid as to encode width%16!=0...
> 
> Really? 640x360 is still more popular than 640x352, I guess...

huh? i rarely see 640x360 files, if ever...

> DRFI>> There's your problem. After a little fiddling, I can reproduce your
> DRFI>> bug with:
> DRFI>> video resolution not divisible by 16,
> DRFI>
> DRFI>704 _is_ divisible by 16 though. And....if Vladimir is such a
> DRFI>stickler for quality, I can't imagine he would encode with a bad
> DRFI>width...
> 
> I do. Actually, I encoded with width divisible only by 4 in the past,

lol. i hope you saw the figures. 5-10% higher bitrate required to get
same psnr!!

> DRFI>> BTW, multiples of 16 (not just 8) really are a good idea, and are very 
> DRFI>> cheap compared to everything else you could do to improve video quality: 
> DRFI>> At a given PSNR (and I see no reason why this comparison shouldn't be 
> DRFI>> adequately measured by PSNR), A mpeg4 video with both dimensions 
>                                             note this  ^^^^^^
> DRFI>> non-mod16 takes 5-10% higher bitrate than one cropped, downscaled, or
>                                        and this ^^^^^^^^^^^^^^
> DRFI>> even upscaled to the nearest mod16.
> DRFI>> The penalty is about half that if only one dimension is non-mod16.
> DRFI>> Even with 4mv and a resolution divisible by 8, you don't get much of 
> DRFI>> that back (though it depends on the content).
> DRFI>> Maybe I'll get around to posting graphs/writeups of my mod16 benchmarks 
> DRFI>> sometime...
> DRFI>
> DRFI>Hahaha, wow, I'm glad to finally have some quantitative figures on
> DRFI>this. The 5-10% figure is huge...I'd like to see some info about test
> DRFI>methodology.
> 
> His methods must be incorrect (well, not really, but they aren't showing
> the real situation, because this is comparsion of the worst case with
> the best case). With the method "encode with qp=2 and compare the
> difference in filesize" I get insignificant difference. Up to 1%, maybe.

your test is not valid. you need to compare the psnr for both files
too. the psnr for the incorrect-sized file will be much worse because
the partial blocks at the edge don't encode well!

> Of course, _both_ files were cropped correctly and usually one of the
> dimensions was multiply of 16. The other was, say, mod8. Or even only
> mod4.

mod4 is useless, and mod8 barely helps. at least you're not stupid
enough to be using non-mod2 dimensions tho (will misalign chroma)...

at this point, you've repeatedly insisted on doing things that are
proven to lower quality significantly, and then you complain about
lavc being low quality without any proof, when all my experience has
shown the opposite. this conversation is getting really stupid. if you
want to continue it, post a sample "challenge" file with your xvid
encode and the original, and i'll encode with lavc.

rich