International organisation for standardisation organisation internationale de normalisation

5.11.3Prediction methods

5.11.3.1.1.1.1.1.1JCTVC-H0401 Differentiated weighting parameter sets for weighted prediction [J. Xu (Sony), A. Tourapis (Magnum), A. Tabatabai (Sony)]
5.11.3.1.1.1.1.1.2JCTVC-H0546 On weighted prediction precision [M. Coban, M. Karczewicz (Qualcomm)]
5.11.3.1.1.1.1.1.3JCTVC-H0643 Cross-check of weighted prediction precision reduction (JCTVC-H0546) [A. Tanizawa, T. Chujoh, T. Yamakage (Toshiba)] [late]

5.12Motion and mode coding

5.12.1.1.1.1.1.1.1JCTVC-H0709 BoG Report on MV Coding [B. Bross]

Contributions relating to motion and mode coding were reviewed in a BoG coordinated by B. Bross. Notes regarding the contributions reviewed and the discussions held in this BoG were included in the BoG report.

The BoG recommended adopting the following:

• 
  Merge estimation region syntax (JCTVC-H0082)
  
• 
  One merge candidate list for all partitions inside a 8x8 CU (JCTVC-H0240 Variant 2 conditioned on log2_parallel_merge_level_minus2 > 0 as described in H0082 section 6)
  
• 
  Setting the merge TMVP refidx to 0 for the non-first partition (JCTVC-H0278 / JCTVC-H0199)
  
• 
  Removing non-scaled bi-predictive merging candidates (JCTVC-H0250 / JCTVC-H0164)
  
• 
  Clipping scaled MV to 16 bit and adjust it according to profile/level decisions (JCTVC-H0216 / JCTVC-H0555)
  
• 
  Removing the list empty check and the duplicate check in AMVP for zero motion vector (JCTVC-H0239)
  
• 
  Removing redundant spatial candidates check in AMVP (JCTVC-H0316 first part)
  
• 
  Fix AMVP non-scaled/scaled candidate WD/HM mismatch by correcting the WD (JCTVC-H0462)
  
• 
  Motion prediction at entropy slice boundary (JCTVC-H0362)
  
• 
  Limiting Collocated Temporal Reference to One Per Picture (JCTVC-H0442)
  
• 
  Modification of bi-prediction syntax CE9 BP08 (JCTVC-H0111)
  

Further investigation was recommended on

• 
  Improvement to parallel merge estimation JCTVC-H0082 (JCTVC-H0090)
  
• 
  Effect of Setting the merge TMVP refidx to 0 for the non-first partition on NxN partitions (JCTVC-H0278, JCTVC-H0092, JCTVC-H0214, JCTVC-H0430)
  
• 
  CU-based parallel inter mode (JCTVC-H0279)
  
• 
  Slice level enable_temporal_mvp_flag and marking process (JCTVC-H0570)
  
• 
  AMVP non-scaled/scaled candidate WD/HM mismatch by changing HM (JCTVC-H0462)
  

In the discussion of the BoG report in Track A, regarding H0082 and H0090, it was agreed to adopt H0082, but reluctance was generally expressed regarding the potential for H0090 due to the desire to keep the decoder impact to a minimum for enabling this encoder parallelism.

There was also some Track A discussion of H0111 and H0431 in the context of discussion of the BoG report, although the group ultimately decided to agree with the BoG recommendation in that regard. Remarks of that discussion included the following:

• 
  H0111 is also a customization for a particular type of encoding, which is the type of encoding done by the HM in the context of uni-directional biprediction. More extensive searching in the HM context could give 0.7% improvement of GPB (with 30% increase in runtime), and encoder-only modification without more extensive searching could give 0.2%, and the proposed modification give 1% – all in the LB context. To some extent the question may be how important we consider the LB usage case to be.
  
• 
  Another proposal H0431 suggested to separate the context models for the zero value of the motion vector differences into three cases: uniprediction, List 0 in bipred, and List 1 in bipred. It seems to have about the same gain when used the same way, but appears to be a more general approach rather than a customization to one specific encoding method which always sets the list 1 motion vector to 0 (by measuring statistics rather than forcing one value all the time). The text for this proposal had some problems.
  

It was indicated that text was nearly fully developed for the actions recommended in the BoG report, and that such text would be uploaded as a revision of the report. Decision: Adopt BoG recommendations as described above.

Regarding further investigation, the BoG had requested for the JCT-VC to decide whether a continuation of the prior CE9 should be established versus having all further study should be performed in an AHG.

Although further study of some items was encouraged, there seemed to be no need for a CE on this topic for the next meeting cycle.