On Mon, Mar 28, 2005 at 05:02:18PM +0200, Jukka Tastula wrote:
2500 is barely enough for an svcd. You can't seriously expect that to be enough for 10 times the pixels, can you?
I can't agree with you about this point, I use Q=1000 for pal resolution of VDR reccordings (3 pass mencoder) with thoses options :
-vf crop=$2 -sws 2 -oac copy -ovc lavc -lavcopts vcodec=mpeg4:vbitrate=$Q:trell:mv0:mbd=2:v4mv:qprd:cmp=10:subcmp=10:mbcmp=10:predia=2:dia=2:vpass=1:turbo
Doing the same with Q=2500 was not that bad on my source, but really not perfect...
Perfect quality? mencoder -oac copy -ovc lavc -lavcopts \ vcodec=mpeg4:vqscale=2:keyint=1:autoaspect (:ilme:ildct ) input -o output
That'll make a big file though, probably bigger than the original, but it'll be pretty much as good as it can get. Of course it'll be worse than the original, lossy compression and everything. You probably wont notice. ilme:ildct is probably a good idea if the source is interlaced.
I'm not at all certain if mbd=2 and qns make any difference when encoding with constant quantizer. Qns takes way too much time anyway at that resolution so just forget about it.
Obviously this isn't very useful. It'd be a much better idea to specify an average bitrate and do a two or three pass encode. That way you can control how big the output file gets. Mplayer docs describe the process very well.
I never thought of anything else than 3 passes :-)