[MPlayer-users] Obscure bug in -vo null

[D Richard Felker III]

On Thu, Oct 21, 2004 at 07:09:18PM -0700, RC wrote:
> Okay, here's a strange and obscure bug for everyone to look at...  First
> off, I'm currently using MPlayer dev-CVS-041008-12:19-3.2.2
> 
> This seems to be a bug in -vo null, when used in combination with
> 'cropdetect' AND 'scale'.  It might have other effects, but that's the
> one I've come across.  There is an easy work-around, and that is using
> -vo md5sum instead.  So, there's no big hurry to fix it, but it just
> takes 2 or 3 of these little weird bugs to make using mplayer terribly
> frustrating.
> 
> Let's try this simple case for crop-detection:
> 
> mplayer -vf pp=md,scale=-2:480,cropdetect -frames 10 -ss 120 Farscape\
> 2004Oct18_01\:00.01.mpeg -vo null
> 
> 
> SwScaler: BICUBIC scaler, from Planar YV12 to Planar 444P using MMX2
> VO: [null] 640x480 => 640x480 Planar 444P  [fs]
> crop area: X: 1..602  Y: 51..479  (-vf crop=592:416:6:58)%  0.0% 0 0 70%
> crop area: X: 1..602  Y: 51..479  (-vf crop=592:416:6:58)%  0.0% 0 0 70%
> crop area: X: 1..602  Y: 51..479  (-vf crop=592:416:6:58)%  0.0% 0 0 93%
> crop area: X: 1..637  Y: 51..479  (-vf crop=624:416:8:58)%  0.0% 0 0 93%
> crop area: X: 1..637  Y: 51..479  (-vf crop=624:416:8:58)%  0.0% 0 0 92%
> A: 411.0 V: 411.0 A-V:  0.012 ct:  0.124   11/ 11   0%  0%  0.0% 0 0 92%
> 
> Exiting... (End of file)
> 
> Well, I can tell you those aren't the dimentions I'm looking for... 
> That crops the border off the top, but leaves the border across the
> bottom.  Changing only the -vo gives the correct results:
> 
> mplayer -vf pp=md,scale=-2:480,cropdetect -frames 10 -ss 120 Farscape\
> 2004Oct18_01\:00.01.mpeg -vo md5sum
> 
> SwScaler: BICUBIC scaler, from Planar YV12 to Planar YV12 using MMX2
> VO: [md5sum] 640x480 => 640x480 Planar YV12  [fs]
> crop area: X: 3..556  Y: 52..402  (-vf crop=544:336:8:60)%  0.0% 0 0 93%
> crop area: X: 3..556  Y: 52..403  (-vf crop=544:336:8:60)%  0.0% 0 0 93%
> crop area: X: 3..556  Y: 52..405  (-vf crop=544:352:8:52)%  0.0% 0 0 93%
> crop area: X: 3..632  Y: 52..411  (-vf crop=624:352:6:56)%  0.0% 0 0 92%
> crop area: X: 3..632  Y: 52..411  (-vf crop=624:352:6:56)%  0.0% 0 0 92%
> A: 411.0 V: 411.0 A-V:  0.012 ct:  0.124   11/ 11   0%  0%  0.0% 0 0 92%
> 
> Exiting... (End of file)
> 
> I have to assume the problem is something in the conversion to 444P that
> happens if I use -vo null.  This isn't a problem if I don't put
> scale=-2:480 before cropdetect, but it is necessary.

i don't see the bug right off. one workaround of course is to use
format=yv12 filter after scale in the chain.

my guess is either that vf_scale is broken for 444p output, or
vf_cropdetect is broken for 444p input. but i can't find any bugs in
cropdetect..

rich