JVET-AF0280 EE2-related: complete results for EE2-1.1b in JVET-AF0249 [T. Dumas, F. Le Léannec, R. Utida (InterDigital)] [late]
During the 31st JVET meeting, Geneva, 11-19 July 2023, the tests of ECM with various maximum MTT depth increments were requested, c.f. EE2-1.1b in JVET-AE2024. This will provide a fair comparison for EE2-1.1a evaluating ECM with the non-square QT, c.f. JVET-AF0108. JVET-AF0249 reports the cross-check of JVET-AF0108. Besides, JVET-AF0249 gives the final mean BD-rate versus encoding runtime curve resulting from the runs of ECM with different maximum MTT depth increments. This contribution reports all the results used to construct this curve.
It is reported that, in RA configuration, over all classes, at equivalent encoding runtimes, ECM including the non-square QT yields -0.04% of mean BD-rate gains in luma with respect to ECM with the maximum MTT depth increased by 1 on I-slices and the maximum MTT depth increased by 1 for the first two TIDs.
This contribution is for information. It was suggested that in the future, contributions with normative changes to partitioning should always bring runtime vs. gain tradeoff curves as shown in this contribution and compare to non-normative encoder-only changes.
It was commented that this is reasonable request and future partitioning contributions should bring such information to JVET.