[Mp4-tech][Video][H.264]Question about MB skip in MBAFF

Thu Jan 6 18:43:09 EST 2005

Zhao et al,
Without spending a lot of time looking at it, I recall that there are some cases where the processes described in the standard seem to operate on smaller blocks, but if you trace through all the logic, the results of multiple small blocks turns out to ordinarily act as if using a larger block.  And there is some reason we specified it that way, although I can't say what they were off the top of my head.  I think there is no new error in the text of the standard being reported here, but I would feel more comfortable if someone very familiar with this topic (e.g., Alexis or Barry or Lowell) could confirm.
One thing I suggest is making sure you're looking at the latest drafted corrigendum-ized version of the standard (JVT-M049d3 applied to JVT-L050d5 or JVT-M050d1).
Best Regards,
Gary Sullivan
________________________________
	From: mp4-tech-bounces lists.mpegif.org [mailto:mp4-tech-bounces lists.mpegif.org] On Behalf Of Zhao Ran
	Sent: Sunday, December 12, 2023 11:59 PM
	To: Gary Sullivan; Alexis Michael Tourapis; Isabelle Perroux; mp4-tech lists.mpegif.org
	Cc: yxie_2000 yahoo.com.cn
	Subject: Re: [Mp4-tech][Video][H.264]Question about MB skip in MBAFF
	It is not a bug report. We are just developing a H.264 Codec and use JM decoder for compliance verification. I see Alexis' comment completely and appreciate your help.
	By the way, I has another confusion about Direct mode.
	In Table 7-11 and Table 7-15 of subclause 7.4.5, MbPartWidth and MbPartHeight are used in the derivation process for mv and reference frame indices in 8.4.1 for direct mode prediction. They are 8 and 8 for B_Direct_16x16 and B_Skip, or 4 and 4 for B_Direct_8x8. 
	While according to 6.4.7.5,
	 - If mb_type is equal to P_Skip or B_Skip, or mb_type is equal to B_8x8 and sub_mb_type[ mbPartIdx ] is equal
	to B_Direct_8x8, predPartWidth = 16.
	NOTE - When sub_mb_type[ mbPartIdx ] is equal to B_Direct_8x8, the predicted motion vector is the predicted
	motion vector for the complete macroblock independent of the value of mbPartIdx.
	Well, then why there are 8 and 4 in Table 7-12 and 7-15 instead of 16. Is there any inferred effect of 8 and 4.
	----- Original Message ----- 
		From: Gary Sullivan <mailto:garysull windows.microsoft.com>  
		To: Alexis Michael Tourapis <mailto:alexismt comcast.net>  ; Zhao Ran <mailto:ranzhao arcsoft.com.cn>  ; Isabelle Perroux <mailto:isabelle.perroux allegrodvt.com>  ; mp4-tech lists.mpegif.org 
		Sent: Monday, December 13, 2023 10:25 AM
		Subject: RE: [Mp4-tech][Video][H.264]Question about MB skip in MBAFF
		Perhaps is Zhao Ran trying to report that our reference software encoder has a bug in this area (using a different mode than the decoder reference software)?
		-G.
________________________________
		From: mp4-tech-bounces lists.mpegif.org on behalf of Alexis Michael Tourapis
		Sent: Sun 12/12/2023 2:49 PM
		To: 'Zhao Ran'; 'Isabelle Perroux'; mp4-tech lists.mpegif.org
		Subject: RE: [Mp4-tech][Video][H.264]Question about MB skip in MBAFF
		Dear Zhao,
		The use of skip macroblocks is basically an encoder's decision. Noone forces you to use them and you can even have an encoder that uses lets say only 16x16 inter macroblocks only. Obviously in the case you are describing skip frame MB pair is not allowed and your encoder should not even examine that particular case within your mode decision. In general it is the encoder's job to ensure that such illegal cases never happen. However you may choose to still code this pair as a frame pair and "emulate" one of the skipped macroblocks using the 16x16 inter mode, with mv parameters equal to that of skip, and cbp=0. 
		Best regards
		Alexis
________________________________
		From: mp4-tech-bounces lists.mpegif.org [mailto:mp4-tech-bounces lists.mpegif.org] On Behalf Of Zhao Ran
		Sent: Sunday, December 12, 2023 4:12 AM
		To: Isabelle Perroux; mp4-tech lists.mpegif.org
		Subject: Re: [Mp4-tech][Video][H.264]Question about MB skip in MBAFF
		Thanks for your information.
		The problem I confronted with is that the skipped macroblock pair is frame MB, while the neighbouring macroblock pair immediately to the left of the skipped macroblock pair is field MB. According to 7.4.4, the mb_field_decoding_flag shall be inferred to be 1, and decoder whould reconstruct the skipped macroblock pair as field macroblock pair. Thus, the output of decoder will be different to the reconstruction result of encoder.
		----- Original Message ----- 
			From: Isabelle Perroux <mailto:isabelle.perroux allegrodvt.com>  
			To: 'Zhao Ran' <mailto:ranzhao arcsoft.com.cn>  ; mp4-tech lists.mpegif.org 
			Sent: Friday, December 10, 2023 8:57 PM
			Subject: RE: [Mp4-tech][Video][H.264]Question about MB skip in MBAFF
			Zhao, 
			Please refer to 7.4.4 :
			When mb_field_decoding_flag is not present for either macroblock of a macroblock pair, the value of mb_field_decoding_flag is derived as follows.
			-      If there is a neighbouring macroblock pair immediately to the left of the current macroblock pair in the same slice, the value of mb_field_decoding_flag shall be inferred to be equal to the value of mb_field_decoding_flag for the neighbouring macroblock pair immediately to the left of the current macroblock pair,
			-      Otherwise, if there is no neighbouring macroblock pair immediately to the left of the current macroblock pair in the same slice and there is a neighbouring macroblock pair immediately above the current macroblock pair in the same slice, the value of mb_field_decoding_flag shall be inferred to be equal to the value of mb_field_decoding_flag for the neighbouring macroblock pair immediately above the current macroblock pair,
			-             Otherwise (there is no neighbouring macroblock pair either immediately to the left or immediately above the current macroblock pair in the same slice), the value of mb_field_decoding_flag shall be inferred to be equal to 0.
			BR
			Isabelle
________________________________
			From: mp4-tech-bounces lists.mpegif.org [mailto:mp4-tech-bounces lists.mpegif.org] On Behalf Of Zhao Ran
			Sent: vendredi 10 décembre 2004 10:59
			To: mp4-tech lists.mpegif.org
			Subject: [Mp4-tech][Video][H.264]Question about MB skip in MBAFF
			Hi, all
			I have a confusion about macroblock skip when MBAFF is used.
			If a field MB pair is skipped followed with a coded frame MB pair, the mb_skip_run should be 2 and decoder should copy data from reference frame or field. My question is how decoder know the skipped 2 macroblocks are field MB or frame MB?
			Thanks in advance.
			Best Regards,
			Zhao Ran
-------------- next part --------------
An HTML attachment was scrubbed...
URL: /pipermail/mp4-tech/attachments/20050106/67de863b/attachment-0001.html