JVET-U0133 BoG report on CE complexity analysis [A. Browne]

This BoG was established with the following mandates:

• To perform more analysis on the impact of different proposals in terms of introducing dependencies between blocks, and possible impact on parallel operation, pipelining and throughput.
• To also include proposals for non-CE and CE-related contributions from section 5.1.3 of the meeting report in this analysis.

The BoG met at the following times during the 21st JVET meeting:

• January 8th 13:00 – 15:00
• January 8th 15:20 – 16:40
• January 11th 13:00 – 13:55

The BoG initially reported back to JVET at 0840 on Tuesday 12 January.

Complexity analysis of CE and CE related had been conducted, but was not finished yet.

New results tables with separate results for 12 and 16 bit for SVT had been provided, to be uploaded as new version of JVET-U0022.

Gains are in the range of 2.5–4.3% for 12 bit in RRC (CE-1.x and CE-3.x), around 1.5% for TSRC (CE-2.x) for 12 bit coding, and around 20–28% for 16 bit.

The BoG met again and reported back at 0720 on Wed 13 January.

The BoG report presents an assessment on the number of operations for proposed modification to the Rice Derivation process (clause 9.3.3.2) for CE1.1, CE1.2.a and CE1.3. These tests feature no inter-block dependencies.

It was suggested that a fixed shift does not count as an operation in these assessments and can be considered “free”.

Detailed analysis on the individual proposals can be found in the document (>15 pages).

In the discussion following the report, it was suggested to continue the CE for further studying:

• Performance at higher QPs (relevant for consumer applications)
• Interaction between entropy coding modifications and JVET-U0052
• Investigate benefit over HEVC RExt also in low QP range, for 12 and 16 bit

It would be beneficial to use more diversified test material. This could also be HD, which would also simplify the amount of simulation resources.

It should be identified whether a simple modification of the VTM can be agreed as an anchor (at least for the 16 bit case, where something is obviously broken).

A BoG (coordinated by A. Browne) was requeted to propose the plan for the next round of CE, identify which of the current proposals should be further investigated, and discuss modifications of test conditions (sequences, QP ranges, anchor for 12/16 bit, additional HEVC RExt anchor).

JVET-U0133 BoG report on CE complexity analysis [A. Browne]

See the notes for this BoG report in section 5.1.1.