[M4IF Technotes] Mismatch of IDCT algorithms

[Kristine N. Kneib, Ph.D.]

Good Afternoon Gary and Christoph and MPEG4 Colleagues,
This IDCT mismatch condition of intra refresh every 132 MBs is a piece of 
ancient history (harking back to 1984/5 and H.320/H.261 standards 
development days). H.263 has compliance requirements with H.261 and MPEG4 
has H.263 as a subset. So, old "Grandaddy" H.261 and its compliance is 
still in play in the latest and greatest MPEG4 video coding.
The original IDCT mismatch specs of H.261 defined 9 tests and their 
results. These have been enhanced and refined - sometimes very subtlely - 
over time through MPEG-1, MPEG-2 and MPEG-4. The particular one that is 
referenced here reflects into the numerical accuracies in the data word 
lengths in the multiplier for the IDCT for 8-bit video. Some sequences will 
catastrophically fail if this is not met.
I firmly agree with Gary's answers to the first two points.
Best regards from a long-time H-DOT and MPEG product activist!
Kristine Kneib
KNK Seminars & Strategies
Making MPEG Work for You
www.knk-mpeg.com
  At 12:02 PM 1/21/02 -0800, you wrote:
>Christoph,
>
>Klaus Diepold was also discussing IDCT mismatch problems recently.
>You might want to check with him.
>
>+> 1.) To which extend is the intra refresh every 132 frames required ?
>
>I believe it is required.
>
>+> 2.) Can an MPEG-4 bitstream exceeding the 132-refresh-rule actually
>be
>+>     valid ?
>
>I believe the answer is No.
>
>+> 3.) Does the MoMoSys reference codec really ignore this refresh,
>+>     and if so, why ?
>
>I don't know.  Could be a bug.
>
>Best Regards,
>
>Gary Sullivan
>
>+> -----Original Message-----
>+> From: Christoph Stadler [mailto:Christoph.Stadler   dynapel.de]
>+> Sent: Monday, January 21, 2024 7:40 AM
>+> To: technotes   lists.m4if.org
>+> Subject: [M4IF Technotes] Mismatch of IDCT algorithms
>+>
>+>
>+>
>+> Dear MPEG-4 community,
>+>
>+> I am currently concerned with the accumulation of mismatch errors
>+> when repeatedly adding residual images during MPEG-4 video decoding.
>+>
>+> The MPEG-4 Visual standard (e.g. document N4350 ISO/IEC 14496-2) says
>+> that every macroblock is required to be 'intra' coded at least every
>+> 132 reference VOPs.
>+> But this requirement is somewhat hidden as 'Note 1' of the
>+> (normative)
>+> 'Appendix A'.
>+> And the MoMuSys reference codec seems to ignore this rule too.
>+>
>+> I understand that IEEE 1180-1990 demands a limitation of the
>+> number of
>+> residuals that may be added even with a conform IDCT. And our tests
>+> show, that violation of this can cause a severe degradation in image
>+> quality if encoder and decoder use different (but conform) IDCT
>+> algorithms.
>+>
>+> So there are my questions :
>+> 1.) To which extend is the intra refresh every 132 frames required ?
>+> 2.) Can an MPEG-4 bitstream exceeding the 132-refresh-rule
>+> actually be
>+>     valid ?
>+> 3.) Does the MoMoSys reference codec really ignore this refresh,
>+>     and if so, why ?
>+>
>+> Thanks in advance and
>+> Best regards,
>+>
>+>         Christoph Stadler
>+> _______________________________________________
>+> Technotes mailing list
>+> Technotes   lists.m4if.org
>+> http://lists.m4if.org/mailman/listinfo/technotes
>+>
>_______________________________________________
>Technotes mailing list
>Technotes   lists.m4if.org
>http://lists.m4if.org/mailman/listinfo/technotes

~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Kristine N. Kneib, Ph.D. - KNK Seminars & Strategies
Making MPEG Work for You!
6333 La Jolla Blvd., Ste. 376 - La Jolla - CA 92037-6622
Tel: 858-459-8058 - Fax: 858-459-3654
www.knk-mpeg.com
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-------------- next part --------------
An HTML attachment was scrubbed...
URL: /pipermail/mp4-tech/attachments/20020121/b862e463/attachment.html