Of itu-t sg16 wp3 and iso/iec jtc1/SC29/WG11

JCTVC-O0352 BoG report on Range Extensions topics [D. Flynn, C. Rosewarne]

BoG review in main session Wed 30th (GJS & D. Flynn).

Decision: Endorsed the following BoG recommendations:

• 
  On RDPCM:
  
  • 
    Unify the ordering of inverse transform and inverse RDPCM operations for inter and intra, and to pick one of the two possible orderings of inverse transform and inverse RDPCM:
    
    • 
      invTS → invRDPCM (this one selected in main meeting review; this aspect chaired by D. Flynn)
      
    • 
      invRDPCM → invTS
      
  • 
    Clean up software implementation of RDPCM (Various contributions have been provided, leave to software co-ordinator's discretion as to how).
    
• 
  Disable Rotation for non-Intra (ie, Inter and IntraBC) blocks with a skipped transform [O0186]
  
• 
  RCE2, change the value of the maximum Rice parameter chosen for the RCE.A1 adoption, as current value (9) is best only for 16 bit. Initialize to 7 instead.
  
• 
  Spec issues:
  
  • 
    Fix an incompatibility with version 1, caused by an intra filtering process being disabled conditional on cu_transquant_bypass_flag. Do this only if the SPS level intra_rdpcm_enable_flag is equal to 1.
    
• 
  On IntraBC:
  
  • 
    Enable Inter RDPCM (whatever method) for intraBC (ie, match the software) [O0185]
    
  • 
    Rename the flags intra and inter RDPCM enable flags to not be so confusing [O0185]
    
  • 
    Only use a single context for the intra_bc_flag (ie, no up/left contextualization) [O0073]
    
  • 
    Define the CIP behaviour for intraBC blocks such that an IntraBC block is regarded as available for determination of intra prediction reference samples, and add an informative note suggesting that an encoder should ordinarily avoid referencing reconstructed pixels from inter blocks when encoding an IntraBC block. [O0155]
    
  • 
    Modify the fast encoder mode decision for IntraBC (software coordinator has discretion to evaluate appropriateness for inclusion). [O0245 (2)].
    
  • 
    Not to change the current intended intraBC reference sample restriction on the neighbouring CTU (see below for clarification), but that there should be some investigation into the cost of various restriction sizes. [discussion of O0074]
    
  • 
    BV coding should use a delta BV based on the previously decoded BV in the CTU (resets at the start of each CTU) [O0122]
    
  • 
    Always use DST for 4x4 Luma TBs with IntraBC [O0183]
    
  • 
    Adopt spec fixes on what samples may be used as reference for intraBC [O0183]
    
    • 
      Don't reference anything outside current CTU + Left CTU,
      
    • 
      Don't reference anything outside current slice/tile,
      
    • 
      Don't reference anything within the current CU being reconstructed.
      
  • 
    Prohibit IntraBC from referencing samples outside the current picture [discussion of O0183]
    

Two methods (JCTVC-O0351 and JCTVC-O0205) were presented for the first time that improve the performance of intra block copying. Further study of these methods is strongly encouraged, possibly in the context of a CE.

If a CE is to be performed, the following methods would be of interest to test the benefits of finer granularity prediction techniques along with padding and vector range restrictions:

• 
  Anchor: with current restrictions
  
• 
  Test: all unavailable samples pre-set to default value (mid grey), overlap search restriction relaxed
  
  • 
    Well defined behaviour (Can't be violated) [JCTVC-O0074]
    
• 
  Test padding: with relaxed vector restrictions for overlap [JCTVC-O0157]
  
  • 
    Performance improvement
    
• 
  Test 2NxN, Nx2N [JCTVC-O0205]
  
  • 
    Performance improvement
    
• 
  Test TU process [no method currently available: maybe if (overlap) {force TU split}] [MS/O0183] (can reference prior TUs within the CU/PU)
  
• 
  Test NxN PU at smallest CU level (follow current split modes for Intra, and each PU has a BV) [QC] (can reference prior PUs within the CU)
  
• 
  Test masking [JCTVC-O0351]
  
• 
  Possible test: relax search in other directions (ie, non-overlapping).
  
• 
  Left search area vs Cost [Comment]
  
• 
  Combinations of the above as appropriate.
  

It was also recommended for a CE to study palette coding. This could be in the context of the above CE, or in another.
A possible CE was suggested relating to Rice parameter initialization.

Further discussion was requested by the BoG for O0132 alpha channel regarding SEI aspects. See notes elsewhere.

[Add notes re additional further study items identified]

Further study was recommended for 4:4:4 DF.