[M4IF Technotes] AC prediction, follow document or follow reference software?

[Yuval Noimark]

We've reported this problem during the last MPEG meeting in Shanghai. The 
decision was to use zero predictors in order to keep consistency with the 
reference software and the VM document.
Yuval.
--------------------------------
Yuval Noimark
IBM Research Lab in Haifa
"Gary Sullivan" <garysull   microsoft.com>
Sent by: technotes-admin   lists.m4if.org
24/10/2023 17:16
        To:     "Lefan Zhong" <Lefan   mediaq.com>, 
<technotes   lists.m4if.org>
        cc: 
        Subject:        RE: [M4IF Technotes] AC prediction, follow 
document or follow reference software?
Lefan, 
That appears to be a bug in the MPEG-4 visual spec.  I just looked too,
and 
I couldn't find a clear statement that non-intra macroblocks
are not used for AC coefficient prediction of Intra MBs, although
I think it would be silly to use them.
I don't recall whether that problem has been
previously reported to MPEG or not.  I will try to do so.
Best Regards,
Gary Sullivan
+> -----Original Message-----
+> From: Lefan Zhong [mailto:Lefan   mediaq.com] 
+> Sent: Wednesday, October 23, 2023 4:25 PM
+> To: technotes   lists.m4if.org
+> Subject: [M4IF Technotes] AC prediction, follow document or 
+> follow reference software?
+> 
+> 
+> Hello folks,
+> 
+> In MPEG-4 document 14496-2, subclause 7.4.3.3, adaptive ac 
+> coefficient prediction, it says:
+> 
+> If the prediction block (block 'A' or block 'C') is outside 
+> of the boundary of VOP or video packet, then all the 
+> prediction coefficients of that block are assumed to be zero.
+> 
+> 
+> Beware that statement doesn't tell us what to do when the 
+> prediction block is 'inter'. Should we use the prediction 
+> block coefficients to do prediction or set to zeros when the 
+> prediction block is 'inter' type?
+> 
+> In the comformance test bitstream Hit007.m4v, 1st P-VOP, the 
+> 2nd macroblock (coordinate (16,0)), the macroblock is an 
+> INTRA block, and AC prediction flag is on, and the 1st 
+> Y-block needs to do AC prediction from the left block. But 
+> left block is an INTER block, the AC coefficients are: 
+> 
+> -1 0 0 -1 0 0 0 0
+> 1 -1 0  0 0 0 0 0
+> 0 -1 0  0 0 0 0 0
+> 0  0 0  0 0 0 0 0
+> 0  0 0  0 0 0 0 0
+> 0  0 0  0 0 0 0 0
+> 0  0 0  0 0 0 0 0
+> 0  0 0  0 0 0 0 0
+> 
+> Should we use
+> 1024
+> 1
+> 0
+> 0
+> 0
+> 0
+> 0
+> 0
+> to do prediction or use
+> 1024
+> 0
+> 0
+> 0
+> 0
+> 0
+> 0
+> 0
+> to do prediction?
+> 
+> In the reference softwares (no matter momusys or microsoft, 
+> no matter version 1 or version 2), they all use zeros to do 
+> prediction in such cases.
+> 
+> For scence of prediction, it is better to use zeros in such 
+> cases. But that is conflict to document.
+> 
+> Thank you for your consideration.
+> 
+> Lefan
+> 
+> _______________________________________________
+> 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
-------------- next part --------------
An HTML attachment was scrubbed...
URL: /pipermail/mp4-tech/attachments/20021101/b137f066/attachment.html