The specs have been simplified (simpler variable length coding, fixed length for some symbols to simplify parsing), and some defaults have been changed.
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Johannes is correct there have been numerous changes in defaults and implementation. For example we have changed from unary coding to a form of exp Golomb coding.This was necessary to allow Dirac to be implemented in hardware (unary coding was impractical in hardware).
More importantly:
PSNR IS NOT THE MOST IMPORTANT FEATURE OF DIRAC.
(excuse my capitals).
I am tired of people focusing so narrowly on PSNR. It is not a particularly good metric of subjective quality. Let me re-phrase that - PSNR is a poor metric of subjective quality. So 0.1dB change in PSNR may mean little, even if true, bearing in mind changes in default values etc. More importantly - what do the pictures look like?
But more than the issue of the picture qulaity is the codec as a whole. How complex is it? Is it simple to understand? Can it be implemented in hardware? How many resources does it take to implement? Does it infringe IPR? Overall all these, and other, issues have a much greater bearing on whether a codec will be successful that a very narrow focus on PSNR.
In Dirac we are trying to create a flexible, veratile codec that can address a range of applications in a simple and efficient manner without payment of royalites. I believe it has the potential to out perform H264, but to do this more work is needed on the encoder side to take advantages of the features that are already in the stream syntax.
So please don't focus narrowly on PSNR in the current verrsion of Dirac, which is , after all, still alpha software.
Regards
Tim
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
There is PSNR loss in Dirac Version 0.6 compared with 0.5.4 for about 0.1dB in average. Any tradeoff for that ?
MYO TUN
The specs have been simplified (simpler variable length coding, fixed length for some symbols to simplify parsing), and some defaults have been changed.
Johannes is correct there have been numerous changes in defaults and implementation. For example we have changed from unary coding to a form of exp Golomb coding.This was necessary to allow Dirac to be implemented in hardware (unary coding was impractical in hardware).
More importantly:
PSNR IS NOT THE MOST IMPORTANT FEATURE OF DIRAC.
(excuse my capitals).
I am tired of people focusing so narrowly on PSNR. It is not a particularly good metric of subjective quality. Let me re-phrase that - PSNR is a poor metric of subjective quality. So 0.1dB change in PSNR may mean little, even if true, bearing in mind changes in default values etc. More importantly - what do the pictures look like?
But more than the issue of the picture qulaity is the codec as a whole. How complex is it? Is it simple to understand? Can it be implemented in hardware? How many resources does it take to implement? Does it infringe IPR? Overall all these, and other, issues have a much greater bearing on whether a codec will be successful that a very narrow focus on PSNR.
In Dirac we are trying to create a flexible, veratile codec that can address a range of applications in a simple and efficient manner without payment of royalites. I believe it has the potential to out perform H264, but to do this more work is needed on the encoder side to take advantages of the features that are already in the stream syntax.
So please don't focus narrowly on PSNR in the current verrsion of Dirac, which is , after all, still alpha software.
Regards
Tim