[Mp4-tech] Why I or B frame can have one fixed QP?

Wed May 10 13:22:07 ESTEDT 2006

Dear Alexis,
It is a very precise explanation of the RC algorithm in JM.
We have seen these limitations and this will become very adverse where the system can not afford the time required to carry RD On mode.
If the system needs to take a decision based on heuristics and may be based
on cost matrics such as SAD, MAD ...., then the decision of Qp based on
RD curve will result in wrong selection of Qp value using the present RC algorithm.
It seems that the RC works okay for Baseline profile.
In my view for Frame level rate control it may result in to moderate results but in case
of MB level (where the Qp changes per MB) will result in more artifacts.
I agree with your statement that this RC is better to be treated as educational 
module. Also for the Constant Bit rate case. For Variable bit rate (i am not saying Constant Qp) still seems to be far off.
In my opinion, the TM5 core model with necessary modifications is better way 
to have practical RC. The only variant is to have own cost function to decide
the Distortation.
Please suggest your view.
Regards
Kaustubh
----- Original Message -----
From: "Tourapis, Alexis" <Alexis.Tourapis dolby.net>
Date: Wednesday, May 10, 2024 0:29 am
Subject: RE: RE: [Mp4-tech] Why I or B frame can have one fixed QP?
> Dear Kaustubh,
> 
> I would suggest checking the function 
> updateQuantizationParameter() and carefully read the code for the 
> case that:
> 
> if(img->BasicUnit!=img->Frame_Total_Number_MB)
> 
> Correct me if I am wrong (I never use this RC myself while the 
> truth is that have been slow in using/updating the JM lately) but 
> you will observe that for I slices the QP remains unchanged (i.e. 
> m_Qc=MyInitialQp), while for B slices the QP depends on the QPs of 
> the last two "reference" pictures (I have not tested this but my 
> guess is that would be the last QP used in those pictures). Note 
> that this rate control obviously does not support hierarchical 
> coding structures, picture level RDO decisions (these are based on 
> multiple encodes and the software does not handle this correctly 
> for RC), and interlace type of encodings. I would strongly caution 
> anyone myself in using this particular rate control especially for 
> evaluation purposes. It does not adapt its statistics very well 
> (and does that quite slowly), can create severe artifacts (do not 
> be misled by RD curves who tend to average results and hide 
> quality variations), but also (if I recall correctly) has issues 
> with high motion sequences. In any case, this is just my personal 
> opinion. I would mainly see the RC in the software as an 
> educational module than a practical one.
> 
> Best regards,
> 
> Alexis
> 
>  _____  
> 
> From: mp4-tech-bounces lists.mpegif.org [mailto:mp4-tech-
> bounces lists.mpegif.org] On Behalf Of kaustubh.patankar vsnl.net
> Sent: Tuesday, May 09, 2024 1:10 AM
> To: Gary Sullivan
> Cc: James (Jianpeng) Dong; mp4-tech lists.mpegif.org; mp4-tech-
> bounces lists.mpegif.org; Tourapis, Alexis
> Subject: Re: RE: [Mp4-tech] Why I or B frame can have one fixed QP?
> 
> 
> But I'm under the strong impression that the current JM rate 
> control algorithm, although it does have some shortcomings, does 
> have the ability to change QP within an I or B picture -doesn't 
> it?  (One of these days we need to improve that rate control.)
> 
> Best Regards,
> 
> -Gary Sullivan
> 
> 
>  _____  
> 
> 	From: Tourapis, Alexis [mailto:Alexis.Tourapis dolby.net] 
> 	Sent: Sunday, May 07, 2024 11:17 AM
> 	To: Gary Sullivan; James (Jianpeng) Dong; mp4-tech-
> bounces lists.mpegif.org; mp4-tech lists.mpegif.org
> 	Subject: RE: [Mp4-tech] Why I or B frame can have one fixed QP?
> 	
> 	
> 	Dear James,
>         
> 	There is no quantization limitation in H.264/AVC. I am guessing 
> your observation relates to the current Rate Control algorithm 
> supported by the JM software. This should only be seen as an 
> example design.
>         
> 	Best regards,
>         
> 	Alexis 
> 
>  _____  
> 
> 	From: mp4-tech-bounces lists.mpegif.org [mailto:mp4-tech-
> bounces lists.mpegif.org] On Behalf Of Gary Sullivan
> 	Sent: Saturday, May 06, 2024 8:31 PM
> 	To: James (Jianpeng) Dong; mp4-tech-bounces lists.mpegif.org; mp4-
> tech lists.mpegif.org	Subject: RE: [Mp4-tech] Why I or B frame can 
> have one fixed QP?
> 	
> 	
> 	I don't think it's true that an I or B frame can only have one 
> fixed QP value in the JM reference software.  No, there is no 
> requirement of fixed QP in the standard.  QP can change (from any 
> value to any other value) on a macroblock (16x16) basis in the 
> standard.         
> 	Best Regards,
>         
> 	Gary Sullivan
> 
> 
>  _____  
> 
>        	From: mp4-tech-bounces lists.mpegif.org [mailto:mp4-tech-
> bounces lists.mpegif.org] On Behalf Of James (Jianpeng) Dong
>        	Sent: Friday, May 05, 2024 4:43 PM
>        	To: mp4-tech-bounces lists.mpegif.org; mp4-
> tech lists.mpegif.org        	Cc: jianpengdong yahoo.com
>        	Subject: [Mp4-tech] Why I or B frame can have one fixed QP?
>        	
>        	
>        	Dear experts, 
>                 
>        	Just wondering why I or B frame can only have one fixed 
> QP in H.264 JM reference software? Is a fixed QP required by 
> H.264/AVC standard? 
>                 
>        	Thanks,
>                 
>        	James
> 
>        	
>  _____  
> 
>        	Yahoo! Messenger with Voice. Make PC-to-Phone Calls 
> <http://us.rd.yahoo.com/mail_us/taglines/postman1/*http://us.rd.yahoo.com/evt=39663/*http://voice.yahoo.com>  to the US (and 30+ countries) for 2¢/min or less.
> 
> 	
>  _____  
> 
> 
> 	
> 
> 
> 	
> 	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.
> 	
> 
>