[Mp4-tech] Questions on H.264-SVC

[adrian guede-lewis]

Dear Experts,
I am currently working on the H.264-SVC codec proposed by the HHI and I was 
wondering if somebody could help me out on a few aspects. I am trying to 
figure out a way to apply timestamps to the scalable output stream for 
efficient interfacing with the transport application. To do so, I need to 
make sure I not wrong on the following points:
- Q1: The encoder performs, for the base layer, a pyramid decomposition 
(UMCTF) in a open loop fashion in conformance with the H.264 AVC syntax. The 
decoder therefore performs an inverse pyramid reconstruction, but this time 
in a close loop fashion since the H.264-AVC standard codec only performs 
close loop decoding (hasn't much choice either!). Is that write ?  Doesn't 
this cause a drift effect?
- Q2: Does the coder perform interlayer prediction on the FGS layers? I 
mean, when decoding, lets say, the CIF resolution, does the QCIF FGS layers 
need to be decoded?
- Q3: How many different types of the interlayer predictions are there? (I 
need to sort out things clearly in order to know what needs to be decoded on 
the base layers). As far as I understood, there’s the Intra  BL prediction 
which does not require inverse MCTF on the base Layers, the Inter Motion 
prediction (using previous spatial layers motion info to code current layer) 
and the residual inter-prediction ( basically consists in just retrieving 
residual data of the corresponding MB in previous layer, is that write ?   
This latter aspect is not clear to me, if somebody could explain it to me, 
it would be great). Are these all the inter-layers prediction tricks? Does 
it mean I do not need to decode completely the previous spatial layers (no 
inverse MCTF or UMCTF required at all)?
- Q4: Finally, I was wondering if these inter-layer predictions were 
applied, for the current spatial layer, using all the previous spatial 
layers (which need to be stored on a GOP basis in this case and would be 
memory consuming) or only based on the previous spatial layer ?
I hope this not too much asking at once,
I would be very grateful if somebody could help me.
Best regards,
Adrian
(Simecom, France)