Organisation internationale de normalisation

5.8MVC profiles

5.8.1.1.1JVT-Z028 ( Prop Profiles) [B.-M. Jeon (LG), W. S. Shim (Samsung), S. Cho (ETRI), G. H. Park (Kyung Hee U.), P. Pandit (Thomson), Y.-L. Lee (Sejong U.)] About MVC coding tools

The JVT has put two new coding tools into the JMVM: illumination compensation (including loop filtering) and motion skip to improve the coding efficiency in MVC. However, the current JD does not contain either of these new coding tools that are found in the JMVM.
Compared with simulcast H.264/AVC, JMVM 5.0 without the two new coding tools reportedly achieves about a 19% bit rate savings on average, while JMVM 5.0 with two new coding tools reportedly achieves about a 27% bit rate savings on average, when JMVM common test conditions are used for all test sequences except the "Uli" sequence. This contribution recommended that the JVT adopt the two JMVM tools into the JD (and thus presumably into intended profile plans) at this meeting.
A description of some planned MVC service deployments in Korea was presented, including mobile services in particular.
Question: Are the described deployments planning to use CABAC or CAVLC? Response: CAVLC? Follow-up question: But these coding tools have never been tested with CAVLC, so how could we consider specifying to use this untested configuration?
Question: How many views? 3 (autostereoscopic).
Remark: The gain for these tools, when using only 3 views, will be less than what has been measured in our common conditions tests.
Remark: The best thing for us to do to enable the application at this stage is to establish a standard which is as easy to implement as possible, based on existing implementation designs for AVC. Once the application becomes established, it might then begin to make sense to consider a more customized design.
We seem to only be able to specify profiles at this stage (for completion of standardization by July) without MB-level coding tools, considering the lack of testing of CAVLC performance for such coding tools. Agreed.
Remark: It seems worthwhile to still consider longer-term work that could include new coding tools.
The JVT decided to plan AHG activity to include testing of CAVLC performance  making that a mandate of the phase 2 MVC AHG.

5.8.1.1.2JVT-Z047-Q / M15196 (Late Prop 2.0/3.1) [H. Kimata (NTT), H. Nakamura (JVC), T. Itoh (Fujitsu), T. Nomura (Sharp)] Proposal on Profiles for MVC (Multi-view Video Coding)

Supporting allowing presentation of this late contribution:

• 
  A. Vetro
  
• 
  J. Ridge
  
• 
  F. Istiaq
  
• 
  P. Purvin
  

Apology? Was reportedly sent.

Proposes 4 profiles for MVC.

• 
  Profile A: With no inter-view prediction (basically a simulcast profile), up to 16 views
  
• 
  Profile B: No temporal prediction
  Remark: We ordinarily only specify aspects that have a decoder complexity impact.
  
• 
  Profile C: With inter-view and temporal prediction, possibly with temporal and SNR scalability – inter-view prediction only for anchor picturs
  Remark: Restricting inter-view prediction to anchor pictures seems questionable.
  
• 
  Profile D: Withdrawn.
  

The contribution advocated having some constraints on dependency structures. We should study this.

Suggestion: Define one profile with SPS-level switches of features.
Suggestion: Profile should be based on High profile.
Suggestion: Add a third dimension to capability specification: Not just profile & level, but profile & level & number of views.
Remark: Regarding the inclusion of SVC features, it seems like we haven't thought about how MVC and SVC can work together – does the high-level syntax work for such a scenario?
A. Vetro and H. Kimata considered these as break-out discussion subjects and reported back to the JVT. This was further discussed jointly with MPEG as reported elsewhere in this report. See JVT-Z049 and section 7.2 of this report.

5.8.1.1.3JVT-Z049-B (BoG Report) BoG report on MVC profiles

This document provided a report that summarized the results of break-out group activity discussions on MVC profiles during the Antalya meeting.
JVT decision: It was decided at the meeting to target the definition of a single Multiview High profile. Also, some dimensions of the level definition were outlined and discussed. The current draft text that would specify this profile and the level definition were provided.