Re: [Mjpeg-users] Compare yuvscaler and y4mscaler

[Steven M. S. <sm...@2B...>]

On Sun, 4 Jan 2004 ro...@ub... wrote:

> How does its speed compare to yuvscaler when encoding to
> similar quality (it that is possible anyway)? I have been

	I haven't measured it - scaling is the least cpu intense activity
	that's going on so that has never been the bottleneck.

>   yuvscaler -v 0 -O SVCD -n n
> 
> as a filter. What would be the scaling method whose resulting
> quality best aproximates to the yuvscaler call above?

	Hmmm, what does yuvscaler use by default?  I thought it was a simple
	box or resample filter and that you had to specify the better 
	'bicubic' one with the -M BICUBIC option.

> I do not know yet when a chroma space conversion is needed.

	You most likely do not have to worry about it unless you're 
	using use DV data in its original 4:1:1 format.

> needed. Is there a specific situation where chroma space
> conversion is required?

	If you use either ffmpeg or smil2yuv to get 4:1:1 data then eventually
	that needs to be converted to 4:2:0 before going to mpeg2enc.   
	Filtering (such as yuvdenoise for example) is best performed on the
	original data (before it has been converted, resampled, whatever) - thus
	I obtain 4:1:1 data with 'smil2yuv -i 2 inputfile.dv', do whatever
	filtering (yuvmedianfilter, yuvdenoise, etc) is needed and then
	'y4mscaler -O chromass=420_MPEG2' just before the 'mpeg2enc' command.

> Do the presets set a good scaling method? Or should I worry
> about learning them to choose one?

	No, the scaling method still needs to be selected.   A _very_ good one
	is "-S option=sinc:4", but another good one (perhaps better for SVCD
	use) is "-S option=cubicCR"

> $ mplayer -noframedrop -vo yuv4mpeg -nosound -v -osdlevel 0 \
>     the.two.towers.avi -sub the.two.towers.srt -subpos 78 \
>     -vf expand=:504
> $ cat stream.yuv |
>   y4mscaler -O preset=SVCD |
>   mpeg2enc -v 0 -I 0 -s -f 5 -V 230 -S 800 -a 2 -F 1 -n n -4 2 -2 1 \
>     -b 2800 -B 284 -q 6 -K FILE=matrix.txt -R 0 -E -11 \
>     -o the.two.towers.m2v
> 
> But y4mscaler gives the messages:
> 
>    INFO: [y4mscaler] Input Stream Header:
>    INFO: [y4mscaler] <<<   frame size:  672x504 pixels (508032 bytes)
>    INFO: [y4mscaler] <<<   frame rate:  239759/10000 fps (~23.975900)
>    INFO: [y4mscaler] <<<    interlace:  none/progressive
>    INFO: [y4mscaler] <<< sample aspect ratio:  ?:?
> **ERROR: [y4mscaler] Unknown norm; cannot determine SVCD format.

	That's probably because the input frame size doesn't match one of
	the NTSC sizes (Nx480).   x504 is a very strange size, isn't it?
	NTSC would be 640x480 or so  wouldn't it?

	Ah, I see the "-vf expand :504" in the mplayer command.   Why is that
	present?   Perhaps if that was left out things would work better.

	Is the data coming from a DVD?   If so then do not have mplayer 
	expand/zoom the data - leave it as 720x480 and then y4mscaler will
	see that it is a NTSC stream.

	mplayer's yuv4mpeg output does not seem to get the header fields
	right.   That's why the sample aspect shows up as unknown "?:?".

	Is the data really progressive?   If so then the 'interlace: none'
	is correct.  

> Adding the option "-I norm=NTSC" to the y4mscaler command line does
> not help (the error message is the same from y4mscaler).

	That is a bit of a mystery

	Steven Schultz