ippiFilterDeblocking_HorEdge_H263_8u_I : BUG !

marc_ba · ‎06-04-2004

Hello there,

I've been trying to implement Deblocking filter in H.263+ (Annex J) using

- ippiFilterDeblocking_HorEdge_H263_8u_I and

- ippiFilterDeblocking_VerEdge_H263_8u_I

Unfortunately, it really seems that there is a problem with these functions. Actually, I already have a C++ version of deblocking filter that is working fine, and when I replace it by the functions above, I get crazy results. When I replace them by the MPEG4 version, although it's not the same db filter, it's working much better (of course the result is nor the right one because the db filter process is not exaclty the same but I donot get crazy results !)... Does anyone has already faced this problem ? I'm using my custom DLL built with PX, A6 and W7 ...

Thanks a lot

Marc

Message Edited by marc_ba on 06-04-2004 02:21 AM

marc_ba · ‎06-04-2004

Hello again,

A very interesting information that may highlight a serious bug in ippiFilterDeblocking_HorEdge_H263_8u_I and ippiFilterDeblocking_VerEdge_H263_8u_I : when I build my custom dll using only PX optimisations, it's working fine ... But it bugs when I use A6 ...

Thus, it really seeems thatthere's a bug in MMX/SSE2 implementation of H.263 deblocking filtering ...

Thanks to inform me as soon as possible of that issue,

Regards

Marc

Message Edited by marc_ba on 06-04-2004 02:21 AM

Intel_C_Intel · ‎06-04-2004

Hi Marc,

did you try IPP H263decoder sample? Can you reproduce this problem with it?

Regards,
Vladimir

marc_ba · ‎06-04-2004

Hello,

I would have liked, but there is no deblocking filter in H.263 decoder sample ! Anyway, as I told you, it works just fine when using only PX optims, but definitely bugs when using A6 or W7 optims ...

P.S : I precise that my image memory is 64-bit aligned

Message Edited by marc_ba on 06-04-2004 04:33 AM

Message Edited by marc_ba on 06-04-2004 04:43 AM

Intel_C_Intel · ‎06-04-2004

I see, we will check this. By the way, could you upload

may be small part of your bitstream to test?

Regards,
Vladimir

marc_ba · ‎06-04-2004

Thank you very much, this problem is very important to me.

I can upload a (bad) encoded bitstream with db filter, butthe mentionned problemcan be checked easily with an simple example that uses :

1. The PX version

2. The A6/W7 version

Because these versions dont give (at all !) same results with same inputs ...

Edit: here is the reconstructed image on the encoder side using A6/W7 ...

Message Edited by marc_ba on 06-04-2004 05:48 AM

Intel_C_Intel · ‎06-04-2004

Thank you for the report, we will check this functions

Regards,
Vladimir

marc_ba · ‎06-04-2004

Thanks a lot.

Please keep me informedASAP, because this is critical for me.

Regards

Intel_C_Intel · ‎06-04-2004

sure, I'll do.

Intel_C_Intel · ‎06-04-2004

We confirm that it is bug in optimized code in IPP v4.0, this bug will be fixed in the next IPP version

Vladimir

marc_ba · ‎06-04-2004

Thanks ! That's what I call active support !

Do you know when the next version / patch will be released ?

Intel_C_Intel · ‎06-04-2004

Thank you:)

According our policy we can't garantee any particular date for future release. But, in generally,release of the next version of IPP was targeted to the end of the summer.

Regards,
Vladimir