JVET-M0027 CE7: Summary report on quantization and coefficient coding [H. Schwarz, M. Coban, C. Auyeng]
The CE report summarizes the test results and crosschecks reports for CE7 on quantization and coefficient coding. The CE includes 4 tests on transform coefficient coding.
The purpose of this core experiment is to explore the coding efficiency and complexity impact of proposed algorithms for quantization and transform coefficient coding. This core experiment evaluates the impact of:
Context modelling/selection/coding order for syntax elements related to transform coefficients
The objectives of this CE include:
Investigation of the impact of context selection and signalling schemes for transform coefficient levels on the coding efficiency and complexity.
Transform coefficient coding
Tester | Tool | Cross checker | |
7.1 | Tzu-Der Chuang | [1] Constraints on the number of context-coded bins depend on whether luma or chroma is coded (and, for chroma, on the subblock size) | Yu-Chen Sun |
7.2 | Tzu-Der Chuang | [1] Constraints on the number of context-coded bins depend on whether luma or chroma is coded (and, for chroma, on the subblock size) | Jung-Ah Choi (LGE) |
7.3 | Tzu-Der Chuang | [1] Constraints on the number of context-coded bins depend on whether luma or chroma is coded (and, for chroma, on the subblock size) | H. Schwarz (HHI) |
7.4 | Tzu-Der Chuang | [3] Include rem_abs_gt2_flag into first coding pass | H. Schwarz (HHI) |
In JVET-L0145, three methods to constrain the usage of context-coded bins are proposed. The first constraint is dependent on color component and coefficient sub-block size. The second method is to release the constraints values according to the last significant sub-block position to improve the coding efficiency. The third modification is to move the greater than 2 flag into the first pass to further improve the decoding throughput. The details are described as follows.
[1] Different constraints according to colour components and sub-block sizes
The number of context-coded bins (including greater than 2 flag) constraints are set to 30 on 4x4 luma subblocks, 16 on 4x4 chroma subblocks, and 4 on 2x2 chroma subblocks, respectively. The number of context-coded bin constraints on greater than 2 are set to 4 on 4x4 luma subblocks, 2 on 4x4 chroma subblocks, and 1 on 2x2 chroma sub-blocks, respectively.
[2] Last significant subblock position dependent constraint
According to the last significant subblock position, the number of to-be-coded sub-blocks in a TB can be derived. The constraint value is determined according to the color component and the coefficient sub-block size. With the number of to-be-coded sub-blocks (NumToBeCodedSb) and the number of total coefficient sub-blocks in a TB (NumTotalSb), the constraint value (ConstraintValue) is modified as follows.
(if (NumToBeCodedSb * 2 <= NumTotalSb)
( (ConstraintValue = ConstraintValue * 2;
(else if (NumToBeCodedSb * 5 <= NumTotalSb * 4)
( (ConstraintValue = (ConstraintValue * 5) >> 2;
[3] Include greater than 2 flag into the first coding pass
The greater than 2 flag is proposed to be moved to the first coding pass after the parity bit. The parsed greater than 2 flag is used to calculate the locSumAbsPass1in the context modelling of sig_coeff_flag, par_level_flag, rem_abs_gt1_flag, and rem_abs_gt2_flag. The reserved number of context-coded bins for greater than 2 flag in the second coding pass is also merged into the number of context-coded bins for the first coding pass. The test that moving the greater than 2 flag before parity bit will also be evaluated.
Specification of tests:
Test 7.1: Aspect [1] ( (To reduce worst-case context coded bins (2 per coeff to ~1.6)
Test 7.2: Aspects [1] + [2] ( (Improving coding efficiency relative to 7.1
Test 7.3: Aspects [1] + [2] + [3] (Reduces the scan passes relative to 7.2
Test 7.4: Aspect [3] ( (Reduces the scan passes without including other changes
Average test results for CE 7. The table shows results for a high-complexity encoder configurations (same as CTC) as well as a low-complexity encoder configuration (dependent quantization and RDOQ disabled). All results are relative to VTM-3.0 (using the same configuration as the tested approach).
CE 7 | High complexity (CTC) | Low complexity (DQ off, RDOQ off) | |||||||||
Y | U | V | encT | decT | Y | U | V | encT | decT | ||
AI | 7.1 | 0.08% | 0.64% | 0.44% | 102% | 101% | 0.12% | 0.42% | 0.47% | 100% | 101% |
7.2 | 0.06% | 0.19% | 0.10% | 104% | 102% | 0.06% | 0.12% | 0.14% | 100% | 102% | |
7.3 | 0.00% | 0.09% | 0.05% | 103% | 101% | 0.00% | 0.03% | 0.11% | 99% | 101% | |
7.4 | -0.08% | -0.09% | -0.10% | 100% | 101% | -0.05% | -0.08% | -0.08% | 99% | 101% | |
RA | 7.1 | 0.07% | 0.87% | 0.69% | 101% | 100% | 0.09% | 0.45% | 0.58% | 100% | 102% |
7.2 | 0.03% | 0.35% | 0.30% | 101% | 101% | 0.03% | 0.15% | 0.22% | 100% | 102% | |
7.3 | 0.00% | 0.30% | 0.23% | 101% | 102% | 0.02% | 0.14% | 0.17% | 100% | 102% | |
7.4 | -0.06% | 0.03% | -0.06% | 100% | 100% | -0.02% | -0.12% | -0.02% | 100% | 102% | |
LB | 7.1 | 0.09% | 1.58% | 0.89% | 101% | 102% | -0.02% | 1.03% | 0.66% | 100% | 100% |
7.2 | 0.00% | 0.57% | 0.14% | 102% | 101% | -0.01% | 0.40% | 0.13% | 100% | 99% | |
7.3 | 0.02% | 0.58% | -0.12% | 101% | 100% | -0.01% | 0.50% | 0.09% | 100% | 99% | |
7.4 | -0.04% | 0.00% | 0.12% | 100% | 100% | 0.02% | -0.15% | -0.09% | 100% | 99% | |
For test 7.1, there is clear loss, and especially in Class A.
It was asked whether this had been tested with low QPs. It had not.
The current scheme imposes this limit by counting the number of context coded bins in a coefficient group and switching to bypass when a limit is exceeded.
It was commented that limiting the context coded bins at this very low level – the 4x4 level, may not be desirable – that imposing such a limit at a higher level may be better. However, it was commented by others that HEVC has a similar counting and switching requirement.
Aspect number [3] (test CE7.4), which reduces scan passes, was suggested to be “cleaner” and should increase throughput.
Decision (complexity reduction): Adopt CE7.4.
There was interest in further work for reducing the worst-case number of context-coded bins, but the currently tested methods may not be the right approach for that.
A suggestion for the further study was to establish a limit at some higher level rather than on each coefficient group. It was recommended to study the matter of the worst-case number of context-coded bins in an AHG.