Log in

View Full Version : mpeg2 matrix analyzer?


smok3
11th June 2007, 09:13
So as i understand, certain mpeg2 encoders (cce?) can choose the matrix adaptively for every GOP, is there a tool to analyze mpeg2 stream and build me a list of matrices used for each GOP?

Fishman0919
11th June 2007, 10:36
Yep... these won't build a list but can you look.

http://forum.doom9.org/showthread.php?p=938243#post938243

Pick the type of stream you want to look at, MPEG TS, MPEG PS or MPEG Video ES.

DGIndex can build a list, point to the stream and choose under options "Log Quant Matrices". Next, save your project. It will save the log as a .TXT file

smok3
11th June 2007, 11:01
Fishman0919, tnx, i did it, cce is really adaptive and quenc will use a matrix that is called 'mpeg standard' (in hc terms) by default.

is there a .dll for hcenc that would adaptively select the matrix as well? (i mean, did somebody actually did that dll?)

kumi
11th June 2007, 18:06
The LUMGAIN parameter in HC adapts matrices based on scene luminance. But I haven't seen a matrix.dll in the wild, besides the source code example in the HC PDF docs.

Boulder
11th June 2007, 20:35
Supplied in the HC021 package, there are a few different matrix.dlls you can use.

smok3
11th June 2007, 20:48
kumi&Boulder; ic, will do some experimentation with *LUMGAIN vs matrix.dll's :)

kumi
11th June 2007, 21:03
Oops yeah, I forgot about the .DLLs in \misc\adaptive_matrix. They seem to be similar to LUMGAIN in function...

@smok3: Let us know your experiences with LUMGAIN & matrix.dlls please :)

45tripp
12th June 2007, 01:04
they are the lumgain adaptive matrices,
one and the same

smok3
12th June 2007, 11:41
Immersion; do you have the exact definition of what really happens with *LUMGAIN? (i cant find it) - i mean is it just adapting the default matrix or is actually changing between some of the premade ones?

Boulder
12th June 2007, 11:44
See hank315's code snippet in the HC021 release thread.

45tripp
12th June 2007, 18:28
LUMGAIN adapts the base matrix selected (i.e the one you select), based on the luminance of each GOP, using the maths in the code snippet boulder pointed to.