[Mp4-tech] Why MB-skip is not allowed in H.264 for I picture
Tourapis, Alexis
alexis.tourapis dolby.com
Wed May 16 10:22:28 EDT 2007
Dear Shevach,
As you have stated, in some cases this may be useful. However, in many cases this might not. Providing such a mode for example may hurt the performance for many video sequences since that would add additional overhead for every macroblock. Would this overhead be compensated by the presence of a good enough number of intra skip modes within the entire slice to justify its presence? That is difficult to say. And if it does help, how much benefit would it actually provide? Note that such an evaluation would also have to be made for several sequences with different types of spatial "activity" and for different QPs.
Btw, I would also caution you checking only RD results when evaluating skip modes (if of course your question is based on an experiment that you may have already performed) since, if evaluating such modes within a Rate Distortion Optimized codec, their behavior is evaluated as a "thresholding" mode which could be emulated using other means (i.e. considering existing modes while dumping all coefficients). Distortion performance (i.e. PSNR) may also not match the subjective performance of the decoded picture.
Finally, do note that there is already some activity within VCEG where new techniques are presented, including techniques for intra prediction/coding which probably would probably affect the behavior or performance of such a mode. Some of these suggest improvements of even more than 10%. If you are curious about this activity I would suggest checking the following link and examining some of the contributions (check mainly the latest one which is 0704_San)
http://ftp3.itu.ch/av-arch/video-site/
Note, that If you think that this is something useful for future codecs maybe you can propose it within this activity as well.
Best regards,
Alexis
________________________________
From: mp4-tech-bounces lists.mpegif.org [mailto:mp4-tech-bounces lists.mpegif.org] On Behalf Of Shevach Riabtsev
Sent: Wednesday, May 16, 2024 6:00 AM
To: mp4-tech lists.mpegif.org
Subject: RE: [Mp4-tech] Why MB-skip is not allowed in H.264 for I picture
Thanks John
This is an interesting idea to hide the actual I picture in P-picture envelope.
But I see two difficulties:
1 . How to provide IDR pictures and hence random access for a decoder. In other words an encoder has to produce IDR picture at the pre-defined interval for random access.
2. As a decoder sees a skip MB, it derives its MV (in the case when all MBs are Intra, MV=0) and copies the corresponding 16x16 region from the reference picture. In this case a spatial redundancy is not exploited.
In some cases spatial redundancy gives more profit than temporal one, even for Inter pictures. Unfortunately in H.264 standard only Inter SKIP is defined.
Intra Skip MB could have been defined as Intra16x16 MB with CBP=0 and with a prediction mode equal the prediction mode of top or left MB.
Regards
________________________________
From: John Smith [mailto:jsmith911 gmail.com]
Sent: Tuesday, May 15, 2024 9:06 PM
To: Shevach Riabtsev
Subject: RE: [Mp4-tech] Why MB-skip is not allowed in H.264 for I picture
You may encode picture as P, using intra-only modes. There will be some overhead for mb_type but mb_skip will be available for use.
________________________________
From: mp4-tech-bounces lists.mpegif.org [mailto:mp4-tech-bounces lists.mpegif.org] On Behalf Of Shevach Riabtsev
Sent: Tuesday, May 15, 2024 3:50 PM
To: mp4-tech lists.mpegif.org
Subject: [Mp4-tech] Why MB-skip is not allowed in H.264 for I picture
Dear experts
Unlike other standards (H.261, Mpeg2 and Mpeg4), spatial redundancy in H.264 is much more exploited. Indeed, there are 9 intra prediction modes for INTRA4x4, 4 modes for 16x16 versus DC/AC prediction of Mpeg4 or DC only prediction in Mpeg2. Consequently for smooth I-pictures zero coded_block_pattern is frequently observed.
I wonder why mb-skip is forbidden in AVC for I-pictures. Is there any reason?
Regards, Shevach
Broadcom
-----------------------------------------
This message (including any attachments) may contain confidential
information intended for a specific individual and purpose. If you
are not the intended recipient, delete this message. If you are
not the intended recipient, disclosing, copying, distributing, or
taking any action based on this message is strictly prohibited.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: /pipermail/mp4-tech/attachments/20070516/39e628f8/attachment.html
More information about the Mp4-tech
mailing list