[Mp4-tech] PSNR of B frames.

[Gary Sullivan]

Mark (et al),
The basic idea is that if you code a "Frame A" and use that as a
reference for prediction when coding a second "Frame B" and do not use
Frame B to predict anything else, the bits that you spend to improve the
quality of Frame A will not only improve the fidelity of Frame A itself
but will also reduce the number of bits that you need to use to code
Frame B to the desired degree of fidelity.  So those bits provide a
benefit for the coding of both Frame A and Frame B, whereas the bits
that you spend to improve the quality of Frame B only benefit the
fidelity of Frame B itself.  You therefore get more "bang for the buck"
by coding Frame A well than by coding Frame B well.
The same principle applies hierarchically as well.  As you build up the
structure from layer to layer, the "bang for the buck" generally
decreases.
The result is that the average fidelity is best when the pictures that
form the basis of later predictions are coded with relatively high
quality.
All of this is under the control of the encoder, however.  Encoders
don't need to do that if they don't want to.  There is probably some
relatively-easy way to control that aspect in the JSVM software, but I
am not personally sufficiently familiar with it to instruct you how.
Best Regards,
Gary Sullivan
________________________________
	From: mp4-tech-bounces lists.mpegif.org
[mailto:mp4-tech-bounces lists.mpegif.org] On Behalf Of Barzilaij
	Sent: Saturday, October 14, 2023 4:29 AM
	To: mp4-tech lists.mpegif.org
	Subject: [Mp4-tech] PSNR of B frames.
	Dear Experts,
	It turns out in all my simulations (scalable mode) with the
JSVM, that the B frames in the hierarchical B frames structure always
have a higher initial QP and therefore a lower PSNR value than the I or
P frames (e.g. I or P frames have a PSNR of 38 while B frames have a
PSNR of 34).
	During simulation with VQM Software I analyzed different frame
rates with similar bitrates (CIF format -- 7.5, 15 and 30fps on 100,
200, 500 and 1000kbit/s). In general, even for rather slow moving video
content, the highest framerate was preferred over others for any bitrate
setting. It seems that with extra bitrate budget, increasing framerate
is always preferred over increasing PSNR of lower temporal levels. Since
the B frames can be constructed with little data, jerky or unnatural
motion is very easy to combat.
	Currently I am unable to find good encoder settings such that a
lowering the framerate becomes more useful. My question is the
following;
	Why do B frames have a lower QP and is there a way to adjust the
initial QP value of the B frames? More general, is there a way to
increase the PSNR of these frames to the PSNR of the I or P frames.
	Any feedback is appreciated.
	Kind Regards,
	Mark Barzilay
	VQM Software:
http://www.its.bldrdoc.gov/n3/video/vqmsoftware.htm
	If asked, I can send some results or more info about my
research.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: /pipermail/mp4-tech/attachments/20061014/8ce17bc4/attachment.html