JVET-M0857 BoG report on CE3-related intra prediction and mode coding [G. Van der Auwera]
This BoG report was discussed in Track A on Tuesday 1430–1630 (chaired by GJS).
The BoG reviewed related input contributions to Core Experiment 3 on intra prediction and mode coding, and formulated recommendations for consideration by the track (A).
The CE3-related documents were categorized as follows:
- Cross-component prediction (14)
- Luma intra mode coding (8)
- Chroma intra mode coding (6)
- Interpolation of intra reference samples (4)
- PDPC-related (7)
- Various (7)
The BoG met on 11 Jan. 2019 from 9:00am–1:00pm, from 3pm–7pm, and on 12 Jan. 2019, from 9am–1:00pm, 3pm–6pm.
The BoG made the following recommendations:
- Adoptions:
- JVET-M0064: simplification of the current VTM3 division operation used in CCLM modelling; VTM3 includes 2×512 tables (16 bit elements), which is reduced to one table with 16 entries (4 bit elements), and in addition the bit depth of the model slope parameter (alpha) is reduced from 15 bits to 5 bits. No significant coding efficiency impact was reported. Among proposals, this was the simplest approach. Decision: Adopted.
- JVET-M0095 (editorial): third proposed aspect harmonizes the filtering decision for the reference samples for all directional intra modes (VVC draft Table 8-4 value for nTbS=2 entry modified from 20 to 16). Editor action item: Agreed.
- JVET-M0238: linear interpolation that PDPC uses on the secondary boundary for adjacent angular modes is simplified to nearest neighbour. No significant coding efficiency impact was reported. This saves several calculations. Decision: Adopted.
- CE tests:
- Reducing the number of reference samples used to compute the cross-component linear model parameters: JVET-M0108, JVET-M0211, JVET-M0212, JVET-M0219, JVET-M0229, JVET-M0274
- Multi-model LM (MMLM): JVET-M0384 (determine the block size restriction that is required to improve the worst case of MMLM and evaluate different formulas to determine the knee-point for deriving the two models)
- Harmonization and simplification of MPM list construction: JVET-M0210, JVET-M0239, JVET-M0295 (method 1 + JVET-M0783 = JVET-M0815), JVET-M0494 (methods 2, 4, 5), JVET-M0528
- Question: Does the scope of this test include MPM list construction of ISP, CIIP?
- Remark: It is proposed to have one software codebase for testing the unification proposals and separate the unification tests from intra search RD effects. The planar prioritization aspects, context changes, can be tested on top of the unification tests in additional tests.
- In Track A review Tuesday afternoon, it was suggested that the main goal should be to have a simplification and harmonization the different ways of how modes are selected. It was noted that CE10 also includes other proposals relating to MPM list construction (and for eliminating MPM usage) relating to CIIP and suggested that issues relating to MPM list construction be considered together in a CE.
- Intra reference sample deblocking: JVET-M0138 (comparing the proposed method with strong intra smoothing method from HEVC in terms of complexity, artefact reduction) – this has a subjective effect – in Track A, it was agreed to plan to put this in the same CE as deblocking, since it has a similar need for subjective evaluation of deblocking artefacts.
- Open issues:
- The BoG recommended to present and discuss following documents related to small block size restrictions in the category of complexity analysis and reduction (section 7): JVET-M0065, JVET-M0169.
- JVET-M0099: The BoG recommended first considering other proposals are presented in the luma mode coding category. In Track A review it was agreed that such an optimization proposal should be deferred to later consideration once the elements of the design that are being optimized become more stable.
- The BoG recommended considering the block size restrictions for PDPC preferably when additional hardware/software experts are present: JVET-M0045, JVET-M0122, JVET-M0238. In Track A review, this was suggested to also apply to JVET-M0814
- JVET-M0358: The BoG recommended testing during the meeting, as additional result (AI and RA conditions; 1 tile/CTU configuration), the “AND condition” version of the reference sample availability conditions and report back during the current meeting. The proposal is that when the prediction mode is DC or planary only use PDPC when the above and left samples are both available. A test was run both for CTC and for a case with 1 CTU = 1 tile, so that there would be many instances of partial availability. The coding efficiency difference was very small, but was suggested to be larger subjectively. For the 1 CTU = 1 tile case, the coding efficiency benefit was 0.1%. This would introduce a different type of processing for planar and DC than what is currently included in the design – currently, these modes don’t need a “PDPC off” path. In CE10 there was a consideration of using planar mode without PDPC but this was not done. In the Track A discussion, it was agreed that the benefit seemed insufficient for adding an additional processing type that is otherwise not in the design (i.e., planar and DC without PDPC), since the 1 CTU = 1 tile case should have shown more benefit if there was a strong need. So no action was taken on this.
In the Track A review, there was a comment that there was an aspect of JVET-M0158 that might need further consideration. There are two 4-tap filters for angular mode intra prediction using the reference samples (one smoother one derived by training and the other a sharper filter derived as DCTIF which is the same filter used for MC interpolation of chroma). This contribution proposed using filter tap values derived from a formula instead of the values currently defined in a table. The contributor said there was a phase inconsistency in the current table of phases. An illustration of a case with a sine wave in the reference samples to show a periodic edge generation phenomenon with the current filters.
Another proposal JVET-M0095 would reduce the number of phases from 32 to 16 (for both luma and chroma). No action had been taken on that, as it was considered a minor issue at this stage.
No difference in coding efficiency was shown. This could be kept in mind in further work, but there didn’t seem to be a need to consider this change at this stage. It was remarked that these proposals may be logical, but we don’t need to consider such a small change before the design is more mature.
JVET-M0857 BoG report on CE3-related intra prediction and mode coding [G. Van der Auwera]
See section 6.3.