Organisation internationale de normalisation

Test A.5.1, in which parsing depends on reconstructed palette index values.

Test A.5.2, in which parsing is independent of reconstructed palette index values.

Note: Run time percentages are inaccurate.

The results for A.5.1 are not significantly better than for A.5.2.

Test A.6 – Binarization for run coding

Proponent: Qualcomm, JCTVC-R0174 part 2, JCTVC-S0039

Crosschecker: Canon, JCTVC-S0072

In this test, exponential Golomb and Golomb Rice binarizations and their truncated versions for the palette run values were tested. Combination of tests A.5 and A.6 were also tested.

There were three discussed combinations with similar performance, all with the same number of contexts and the same binarization:

• 
  A.4.2 (with 5 context coded bins)
  
• 
  A combination of A.5.2 + A.6 (having two sets of results, either with 3 or 5 context coded bins, for which the presented CE results use 3, which is similar to the SCM complexity, but 5 seems better).
  
• 
  A related non-CE late proposal (S0269, which was not yet cross-checked), which contains a combination of A.4, A.5, and A.6 and has very slightly better performance, using 5 context-coded bins.
  

(Further discussion chaired by GJS on Monday 10-20 p.m.)

Cross-check S0289 now available for S0269.

S0173 was a related contribution that was suggested by the BoG to be reviewed in this further discussion.

Some data was shown that appeared to indicate that S0173 would not perform as well as S0269. This data was asked to be uploaded so that we could have a record of what it showed.

Decision: Adopt S0269.