JPEG2000 Performance

Deleted_U_Intel · ‎01-14-2005

Hi,

with IPP 4.0 our performance results for JPEG2000 compression are way behind what other commercial products provide, factor (2-5).

The main areas which seem to consume the performance are wavelet transformation and tile enconding.

Does anyone have similar experiences or any suggestions what we can improve?

Can we expect a dramatic performance increase by using IPP4.1?

Vladimir_Dudnik · ‎01-14-2005

Hi,

could you please point to such commercial products? (It is alsointeresting tosee for their cost, is not it?)

In general, that's true, we are not satisfied with performance we had in IPP v4.0 and even in v4.1, in spite of the fact that it was improved. That time we were focused more on optimization for Itanium than for Pentium family and because of that you can see noticeable performance improvement of IPP JPEG2000 codec on Itanium (andmore supported features set) from IPP v4.0 to IPP v4.1. We continue to work on our implementation of JPEG2000 codec in both direction, extending its functionality and improvement its performance.

So, for the next version of IPP you can expect further improvement in this area.We also have aplan to add Linux version of JPEG2000 sample.

Anyway, it is better to submit your issue through Technical Support channel, so, it will be tracked and you will be notified when any changes in this area will be available.

Regards,

Vladimir

Message Edited by vdudnik on 01-14-2005 10:53 AM

Vladimir_Dudnik · ‎01-22-2005

Ashvins

Fri, 01/21/2005 - 12:50

Thanks Vladimir,

I have posted it here to see, if anyone else has similar observings, or if we do anything totally wrong. Unfortunately it seems, that this is not the case.
We have used as an example a 17MB file which is compressed in 43 seconds by the IPP 4.0 Sample and the same image is compressed in 11 sec by the Aware Codec and in 10 sec by Kakadu.

Do you expect to become much better in the future, or do the other two have any dirty (expert) tricks in their implementation?

Hi,

yes we know that there is some performance issue in our sample. BTW, the IPP functions have good enough optimization, so if you are able to usethem in your codec they can give you a value in performance. I do not think that the problem is in anytricks, we just need to continue our work in this area, and we will be definetely better than we is now in the future versions. That's for sure.

Regards,

Vladimir