JVET-O0888 AHG10: Adaptive Coding Sub-set for encoder optimization [M. Bhat, D Gommelet, J.-M. Thiesse, D. Nicholson (VITEC)] [late]

This contribution introduces an additional flexibility capacity at the encoder level. This proposal tries to optimize the coding cost of specific parameters in case of underuse of the full standard capacity. Two scenarios are considered: underuse due to specific content at a given encoding point and desired underuse of modes due to a specific encoder implementation. While several mechanisms are already available to handle these underuses in previous standards and in current VVC draft, including high level flags and CABAC probability update, it is asserted that higher flexibility might bring a better trade-off for encoder design. The contribution proposes a subset signalization of specific coding modes at PPS or Slice level. This is presented in this contribution for two coding elements: the SAO offsets and the intra coding modes. Three way of sub-set signalling are evaluated in the contribution and it is asserted that early results show a potential to be further investigated.

It is proposed to reduce signalling of parameters for tools, e.g. for the benefit of encoders which do not exercise all modes to reduce the signalling overhead.

Though the concept is generic, it would need to be specifically defined for each tool what these subsets would be. This would have impact on the parsing, though not on the subsequent decoding process.

Examples are given for SAO offset values and Intra mode directions.

As an example, if an encoder restricts its search to only 19 intra modes, the loss is approximately 2%. If it is signalled that less modes are used, the loss goes down to 1.5%. Such benefit does not seem to be large.

For SAO, the gain was marginal, and the encoder runtime reduction is also not large.

The specification of low level tools in terms of parsing becomes more complicated as well.

It may not be beneficial for any encoder – more sophisticated encoders may use a different strategy of testing subsets of modes. On the other hand, it makes decoders and specification of parsing more complicated.

No results on amount of reducing encoding time for the intra case.

Though the general idea is interesting, the data are not complete enough to take action here.

It was suggested to run a CE, but this does not seem to be possible as it would need to employ realistic encoders.

Further study was recommended. More convincing data about the benefit should be provided.