Multipass Transcoder

[freaky_00]

Is there a multipass transcoder which analyizes the picture and optimizes the bitrate to achieve best results. I mean like CCE but the transcoder actually always compares the result with the original source and you can chose between 2 to ? passes! Is there, will there ever or will there never be something like this?

[yuinfo]

http://forum.doom9.org/showthread.php?s=&threadid=63587, specialy last post.

[freaky_00]

Well this doesn't mention multipass. It mentions deep analysis!!!

This is what I mean. DVDShrink would make ALWAYS a deep analysis. Now you could set some options and the PASSES! The first pass DVDShrink would do, will be somehow like it is right now. The next passes DVDShrink would somehow (is this possible?) COMPARE THE QUALITY OF THE TRANSCODE WITH THE ORIGINAL SOURCE!!! Like CCE has its VAF DVDShrink could use an equivalent type and compare the two by Q-Scale and Bitrate as in the robshot guide. Could this be implented? Is it possible?

I think we all agree that DVDShrink is one of the best transcoders. Improving the Quality would make it in my eyes perfect!!!

[dragongodz]

freaky_00 - actually something like that sort of thing could be done, even just using 1 pass.
that is it decodes a frame. first just drop coefficients and call it try1, then from the original frame drop half as many coefficients as try1 and requant a small amount, call this try2. then compare try1 and try2 to the original frame and whichever has the higher PSNR or SSIM or whatever measure you choose to use, then write that frame.

actually pinnacle instantcopy is said to do 2 decodes and 1 write so is probablly doing something similar to that already. of course pinnacle are not going to say exactly what. :)

as for options you could pick, this is transcoding we are talking about and as such the options that could be made selectable is limited. if you were to use the above method then that would be true even more so.

[djan]

It's why Instant Copy takes more time and has better quality to my taste. Isn't ?

[dragongodz]

djan -
even doing what i said above though should be faster than what instantcopy is meant to be(havent tested it since looking at version 7 first demo). but yes the double decoding and quality comparison would increase the time it takes substantially.
also yes for the quality since you are trying to pick the best looking reduction method based on a per frame basis.

you could probably increase the speed by making it per GOP based on first of each picture type. that is see what method is best for first B frame and what best for first P frame and use those for the remaining B and P frames in the GOP. once a new GOP is started retest. that way less than a quarter of frames would be tested but compression method would be tested/updated/changed very often.