[M4IF Technotes] Rate Control in VM - Repost

Arcin Bozkurt arcin lumictech.com
Wed Dec 12 16:48:37 EST 2001


The introduction section (15.1.1) mentions that the SRC scheme can
handle I, P and B pictures and the description section is for single VO
and only handles I and P pictures.
My question is what "sequence" or "segment" corresponds to in the
context of SRC.
VM16.0:
R_r : number of bits remaining for encoding this sequence (or segment)
N_r : number of P frames remaining for encoding.
R_p : bits to be removed from the buffer per picture is calculated as
        R_p = R_r / N_r
which suggests that a sequence is
I P P P P I P P P ....  : 1 I and N_r (in this case 3) P pictures... ( For 
describing the pattern, I am assuming value of N_r is the initial value of 
N_r, because later N_r is decremented after the encoding of each frame)
Are the following deductions correct?
1.
R_s : is the target bitrate for an IPPPPPPPPPPP sequence
2.
R_f: (bits used for the first frame) "first" frame always refers to an I
picture.
3.
T_s: (number of seconds for sequence) This is the parameter that gives
the information on how many P pictures follow an I picture (combined
with the N_s information). This means that this SRC does not change the
number of P pictures between I pictures (unless a frame is dropped).
Therefore, suppose a scene change is detected. This scheme cannot be
used, if the encoder forces an I picture (for quality) instead of
continuing coding P pictures with small QP value (we know that SRC
cannot update that value of QP more than 25% - so a degradation in
quality is expected.)
N_s: distance between encoded frame 4 for 7.5 fps.
With N_s = 4, T_s: 10 (examples from VM) i would have 1 I, 74 P pictures
in my sequence...
Am I right?
Arcin Bozkurt


More information about the Mp4-tech mailing list