Log in

View Full Version : Matrix re-allocation in redistribution mode


Sharc
12th June 2007, 22:14
In standard multipass-VBR mode DVD-RB allows -- under advanced options -- to select different matrices that will dynamically kick in during the encode phase depending on the actual cell bitrate.

When enabling the newly introduced re-distribution, the bitrate swing between segments will typically -- and often significantly -- increase. This would suggest to re-allocate the matrices depending on the new bitrate profile, before starting the multipass encode phase.

On the other hand, the redistribution is based on a fixed Q which in a way stands for "constant quality". This would actually forbid to change the matrix dynamically once the profile has been set, as changing the matrix would spoil the underlying concept of "constant quality".

I understand that
a) DVD-RB in redistribution mode will take the matrix which is selected as "main feature matrix", for the redistribution profile and for the encode.
b) RB-Opt v0.29 accepts for redistribution the matrices that it gets from the "prepare" phase of DVD-RB (in standard multipass VBR mode) and leaves them untouched (unless manually changed).
In no case however the matrices will be re-allocated according to the new profile for the encoding.

So the question is if the final quality could be further optimized if DVD-RB would - before starting with the encode phase - re-allocate the matrices in accordance with the redistributed profile, or if the re-allocation would rather spoil the quality.

I wonder what the experts' opinion is .....

robot1
12th June 2007, 22:20
So the question is if the final quality could be further optimized if DVD-RB would - before starting with the encode phase - re-allocate the matrices in accordance with the redistributed profile, or if the re-allocation would rather spoil the quality.

I wonder what the experts' opinion is .....
I'm not expert... anyway here is my opinion.
If you change the matrix *after* the redistribution, you don't have anymore a sort of costant Q between all the segments.
If the pourpose of redistribution is to have a costant Q, you should not change parameters after the redistribution. And, if the pourpose is to have a sort of "costant quality", you should have the same parameters for all the cells.
My 2 cents.