[Mp4-tech] RE: [jvt-experts] why B frames bitrates must be smaller than P framesbitrates?

Tourapis, Alexis Alexis.Tourapis dolby.net
Sun Nov 19 12:33:08 ESTEDT 2006


Dear Bita,
The Lagrangian costs for motion estimation and Rate Distortion Optimized
mode decision are also different (higher) by default in the reference
software. Their values tend to give higher priority to inter modes and
in particular skip since they tend to give a preference to modes that
result in fewer bits instead of lower distortion. The general assumption
(right or wrong) is that for B slices you can sacrifice a bit more in
terms of quality and have savings on bitrate which would translate to
bigger savings and smaller impact in quality on the overall sequence.
You can change the lagrangian costs either by modifying the software or
by using the explicit lagrangian parameters in the configuration file
(UseExplicitLambdaParams)
Furthermore you may wish to set the following flags 
DisableThresholding = 1
DisableBSkipRDO = 1
The second one can be especially important since it avoids forcing the
test of skip modes in B slices (normally the encoder will test if the
motion vectors corresponding to skip/direct also result in a coded
residual and based on that determine if mode is skip or direct).
Although bipredictive ME could improve the chances of getting
skip/direct modes and of course bi-predictive partitions, I do not
understand your reasoning of disabling this. Are you trying to bias
intra modes during fading transitions? Furthermore, I am also assuming
that you are not using weighted prediction which was introduced in the
standard for this particular reason.
Finally, I am assuming you are talking about the encoder.cfg file in
terms of defined QP values for P and B slices. The reference software
does not specify any such condition in general.
Best regards,
Alexis
________________________________
From: jvt-experts-bounces lists.rwth-aachen.de
[mailto:jvt-experts-bounces lists.rwth-aachen.de] On Behalf Of bita
damghanian
Sent: Sunday, November 19, 2023 12:02 PM
To: jvt-experts lists.rwth-aachen.de
Cc: mp4-tech lists.mpegif.org
Subject: [jvt-experts] why B frames bitrates must be smaller than P
framesbitrates?
Dear all experts,
my experiments show that in fade out sequences(where the frames
luminance decrease in consequetive frames), the P frames macroblocks are
mostly Intra coded.
the case is also true for B frames.
but in long fades where the luminance decrease between two consequetive
frames is so small, the B frames MBs are mostly Skipped. but in this
situation the P frames macroblocks are still mostly Intra coded.   
I don't know why when P frames MBs are intra coded, the B frames MBs are
skip coded.  
I found 3 reasons for this:
1- Bi-predictive ME for B frames and being able to be predicted from
different directions, make B frames be coded more optimum.
    to  omit thit effect, I disabled Bi-predictive ME and defined 1
number of reference frames for B frames.
2- In jm10.2 the QP defind for B frames are greater than the QP defind
for P frames and thus B frames are mostly prefer to be skipprd.
    to  omit thit effect, I defined equal QPs for P and B frames
3-rate control should be off
but still the problem remains:
what are the other factors which make B frames MBs to be more skipped
rather than to be Intra coded like P frames MBs?
in other words, why B frames bitrates must be smaller than P frames
bitrates? what are the parameters in the encoder which force this?
thanks in advance for your attention and help,
Bita Damghanian
-----------------------------------------
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/20061119/462ca76f/attachment.html


More information about the Mp4-tech mailing list