Quote:
Originally Posted by jorel
oh yes telemike,
i knew "roba way" a few months ago
and seems to work very fine,but i never test it.
did you use and got good results?
thread in d9:
http://forum.doom9.org/showthread.ph...0&pagenumber=1
i want to know if have a way to work with tmpgenc too.
Bach is brasilian(like me) and some day i will call him to ask.
|
Hi Jorel,
After reading many obfuscated threads in doom9, I think I finally understood "Roba" (thanks to r6d2's and DdogG's posts) .
From what I understood, you do a file size prediction and then do a CQ-based encoding. If the final size is too big or too small, you feed this CQ encode to a "2nd pass" VBR encode, that will fix the size.
If the CQ pass was *slightly* too big, this 2nd pass file fix would not ruin the CQ "magic", hopefully.
If the CQ pass was too small, this 2nd pass file fix could make it better because it would increase bitrate here and there to fill the disk.
Now, with TMPGEnc this is not easy (possible?) because 2-pass VBR is not two separate jobs like in CCE, but a single, looong job. So you don't have how to "fool" it into using a CQ encode as the first pass for the 2nd pass (at least I don't know how).
Also, with TMPGENc it's not so necessary, because TMPGENc's CQ is a float number while in CCE it's an integer. So our prediction is much "tighter" than with CCE.
I think the closer to RoBa we can get with TMPGEnc current versions is what's discussed here:
http://www.kvcd.net/forum/viewtopic....er=asc&start=0