Help for H.264/AVC scalable video coding

[Schleiereule]

Hello!!
excuse my English it is not very good.
I have seen that there is a forum on the standard H.264/AVC SVC and I desired to ask you a thing, I am trying the packet software JSVM with different encoder configurations.
With the fine-grain scalable configuration I get some values of the PSNR upper of 1 or 2 dB to that coarse-grain scalable configuration.
Does difference seem me a lot of, according to you it is normal?
Am I mistaking anything in the configuration file?

Thanks very much
Goodbye
Schleiereule

[BeyondTheEyes]

Dear Schleiereule, all,

FYI, Fine Grain Scalabity (FGS) functionality has been removed from the SVC specification that is still under development (SVC amendment will be finalized by the next (Geneva) JVT meeting beginning of July 07). JVT decision was based on the conclusion that FGS, in its current design, is somehow over-engineered and would need improvements. BTW, FGS is considered as a potential candidate for Phase 2 of SVC.
Nevertheless, JSVM still incorporates FGS encoding tools. Please note that SVC does specify, among other things, fidelity scalability using either Coarse Grain Scalability (CGS) or Medium Grain Scalability (MGS). CGS may be seen as Spatial Scalability with ratio between layers equal to 1, whereas MGS adds high level syntax that allows frame-level bit-rate adaptation therefore graceful degradation. Moreover, MGS enables reducing drift while still ensuring good rate-distortion results thanks to the "key picture" concept. In a nutshell, I would highly recommend using MGS instead of CGS.
Coming back to the previous post, the PSNR gap you are reporting here may be due to some different reasons :
- unpratical bit-rate setting (eg several Mbit/s for CIF resolutions)
- much higher resulting bit-rate when using FGS compaired to CGS ?
- unfair settings for CGS in comparison with FGS (delta Qp setting, motion vector refinement allowed for FGS and not for CGS, etc.
- ...
Could you please post your configuration file, JSVM version etc. so that you can figure out together what is potentially wrong with your settings ? Please keep in mind that specifying a standard is not only about coding efficiency, but tradefoff and compromise between coding efficiency, decoder complexity, application requirements and so on. These 2 dB improvements may be there for your specific case, but that may drive the price to unfordable levels for the decoder and the application you are targetting.

Kind regards,
BeyondTheEyes

[Schleiereule]

Hi BeyondTheEyes,
thanks for all.
I am making a project for the university "analysis of the video quality in the scalable video coding ".
How software I am using the JSVM 7.0 where the MGS configuration is not present.
My problem is to understand if the difference between the configurations is normal,
on some articles I have seen that it was almost null and that a configuration was better than the other,
however I don't know if it is always so.
As you have said, also I have not found anything that for example explained that FGS is better than the CGS....
For the tests I am using the sequences Crew Ice Harbour and the data are very similar.
These are some configurations
In the fgs configuration I use an only fgs layer with the command line, because the bit-rate are of the order 50Kbps 500Kbps and 2Mbps.

Thanks very much
Goodbye
Schleiereule

[BeyondTheEyes]

Dear Schleiereule,

Looking at your configuration files, several issues appear :

- for CGS, CgsSNRRefinement has to be set equal to 0, otherwise MGS coding is used;
I would kindly suggest that your read more carrefully the JSVM software manual;
- AVC base layer is not the same for the two configurations (different QP settings for layer0);
- the QP settings for CGS does not seem to be right : when encoding an additionnal SNR enhancement layer, it is more approriate to decrease the QP instead of increasing it. FYI, a delta Qp between layer equal to -6 results approximatly in doubling the bit-rate; Basically that corresponds to how FGS internally operates.

Therefore, I would kindly suggest using only one CGS enhancement layer, using deltaQp=-6, so as to enable a fair comparison with FGS, in the scope of a coding efficiency comparison and not a bit-rate adaptation scenario, for which the CGS approach is not appropriate (because one can not (a fortiori gracefully) "truncate" a CGS enhancement layer).

Kind regards,
BeyondTheEyes

PS : for this kind of issues, I would encourage you to post your questions to the official JVT SVC mailing-list (jvt-svc@lists.rwth-aachen.de) to which you may want to register (http://mailman.rwth-aachen.de/mailman/listinfo/jvt-svc)

[Schleiereule]

Hi BeyondTheEyes,

Thank you very much for the information!!
I am using an old version of the JSVM (7.0) where there is not the MGS option ,
for the quantization parameters you are right, to the beginning I have used that values subsequently
I have used the command lines to modify them (in decreasing order), to return in the interval of bit-rate I have to use a ΔQP ≈ 2.
I have to use three quality levels, and therefore three cgs layers to limit the occupation of the memory.
A question, in the cgs configuration for a rate distorsion optimized the parameters QP, MeQPLP and MeQPX should be set to identical values for each cgs layer.
The data go better with the other sequences.


Thank you very much in advance for your time and for the link!!

Goodbye
Schleiereule

[Raza]

Dear Schleiereule and All

i am new to jsvm, can any body help me how can i configure main encoder file for scalable coding mode. i am reading the jsvm software manual, but i am not able to understand, how can i configure? and where can i find the main encoder configuration file.

i will be thankful for your kind help.

Regards

Raza.

[anna_kass]

hi to all I'm new to SVC and JSVM
can any one help me : How can i configure the batch file to make the execution
thanks for help
thanks