[M4IF Technotes] Trouble with Bframe: size goes bigger

[Luca Piccarreta]

The question is a little bit ill-posed. May be you're using less bit,
but the quality is better. And what's the measure for quality?
MAD, PSNR subjective tests?
In my experience, B frames don't give any improvement in coding efficiency
(read: MAD at same bitrate). It may be was true for MPEG2, but new tools,
like 4MV and OBMC, have improved the quality of P compressed frames.
Moreover, one should never use more than 2 or 3 consecutive B, as the
motion estimation algorithms could easily fail. And even if they don't fail,
very large motion vectors consume bits. One could argue that in Direct MBs
(MPEG4 tool) the coding efficiency is very high...
On the other side, BVOPs are useful to minimize mismatches between
different decoders (too many consecutive PVOPs can easily cause visible
differencies between different implementations) and make random access
faster.
I would appreciate other opinions, though.
Luca Piccarreta.
  ----- Original Message ----- 
  From: Francesco Falconi 
  To: technotes   lists.m4if.org 
  Sent: Tuesday, May 28, 2024 2:13 PM
  Subject: [M4IF Technotes] Trouble with Bframe: size goes bigger
  Hello to everyone,
  I hope someone can find two minutes to answer me. 
  I show you this test: they are only 8 frames to compress, (frame base cosing). My test is about advantage in size compression using more bframe than Pframe. 
  The GOP is 8 frame and i try to change the number of P-B frame (fixing numer of Iframe).
  So i change the value of  Motion.PBetweenICount  and Motion.BBetweenPCount 
  These are the results:
        P=    B=
       GOP 
       Size cmp
       fps
        0    6
       I B B B B B B I
       15944
       0,4097
        1    3
       I B B B P B B I
       14804
       1,3095
        3    1
       I B P B P B P I
       14065
       2,4133
        6    0
       I P P P P P P I
       14023
       2,8229
  I cant't understand why adding Bframe the size goes bigger, it could be the opposite (accordin Mpeg Theory). I try to make other test, the same results.
  Could you explain me the reason?
  Thank you vaery much, this is for my university thesis.
  Francesco Falconi
-------------- next part --------------
An HTML attachment was scrubbed...
URL: /pipermail/mp4-tech/attachments/20020528/b7c47fbd/attachment.html