JVET-Q0516 MTS signalling based on last significant coefficient position [X. Xiu, Y.-W. Chen, T.-C. Ma, H.-J. Jhu, X. Wang (Kwai)]
In VVC draft 7, multiple transform selection (MTS) can be enabled when there is non-zero transform coefficient in luma transform block (TB). This is done by checking whether the luma coded-block-flag (CBF) is equal to one. In this contribution, it is proposed to replace such CBF based condition with one based on the position of last significant coefficient. More specifically, the signalling of the MTS index (i.e., mts_idx) is conditioned on whether the position of last significant coefficient is less than one. Since the same condition is also used to determine the signalling of LFNST index lfnst_idx, the proposed change incurs no extra complexity. Compared to VTM-7.0 anchors, the proposed change reportedly leads to BD-rate differences of -0.10% and -0.03% for AI and RA configurations, respectively, with no impacts on encoding and decoding runtimes.
The presentation deck had not been uploaded, and was requested.
JVET-Q0685 is identical, however on top of JVET-Q0055 modification.
The benefit of harmonization is not too obvious, as it is just a part of conditions that invoke LFNST and/or MTS.
In terms of compression, the gain over CTC is relatively small and would not justify that action.
The powerpoint deck has additional non-CTC results where MTS is switched on for inter, the gain of the modification seems to be more relevant in that case. Currently, MTS is disabled for inter, as the runtime increase is not justified by the compression gain. However, those results are not complete.
JVET-Q0516v3 reports complete results (reviewed Thursday 16 January in Track B) on enabling the method with inter MTS.
Current MTS gives 0.06% in RA, 0% in LD, encoder runtime approx. 10%
Proposed method gives 0.26% in RA, 0.27% in LD, same encoder runtime increase.
This would still not be a good tradeoff complexity/compression in terms of enabling it in CTC.
The reason for the gain is that it is somewhat reducing redundant signalling.
The proposal is more similar to how signalling was done earlier in JEM.
The encoder is not changed (confirmed by crosscheckers) – proponents will upload the software such that other experts can also check for that.
Several experts expressed support for this proposal as it has some benefit in compression, there is no complexity concern, and there was no strong opposition.
As an alternative, is was suggested by one expert removing inter MTS. However, there is still a small benefit in compression if an encoder wants to use.
Decision: Adopt JVET-Q0516 (no change in CTC).