JVET-AE0185 AHG12: IntraTMP with neighbouring candidates [K. Naser, F. Le Léannec, A. Robert, F. Galpin (InterDigital)] [late]

In ECM-9.0, IntraTMP search is performed within predefined search area to obtain the best candidate block vectors. This contribution proposes to augment the TM search with the block vectors of neighbouring PU’s that are coded with IBC or IntraTMP. The purpose is to improve the IntraTMP by further including probable candidates during the TM search. On top of ECM-9.0, the following results are obtained:

• AI -0.09% 0.18% 0.17% 100% 100%

Five additional candidates in IntraTMP. Search area is not going beyond IBC search range. Only uses BVs that are already stored in current ECM.

Where does chroma loss for some cases come from? This was not analysed.

Tradeoff is reasonable.

It was requested to provide per-sequence results, and RA results in xls spreadsheet, and an provide an answer to the question on partial chroma losses.

In an update (v3) following further investigations by proponents, it was stated that some negative interaction with DBV was found which was responsible for the inconsistency in chroma losses. Was also sent to cross-checkers, but not completed yet.

Are additional candidates checked in the updated version? Yes, likely more than 5 candidates could be checked as maximum (not exactly known), but it was claimed still search area is not going beyond IBC search range, and only uses BVs that are already stored in current ECM.

Run times of the new method are not reported. It was also commented that the new changes are not sufficiently described in v3.

It was also commented that there might be interaction with some of the new ECM adoptions.

Interest was expressed in the benefit of the method, but considering that this was a very late contribution (in terms of the new elements that were added, which were not documented yet), further study was recommended until the next meeting.