JVET-AM0269 AHG17 ECM/VTM performance under CfE requirement [K. Naser, E. François, F. Le Léannec, F. Galpin, T. Poirier (InterDigital)] [late]

During the discussion about CfE test conditions, it was agreed in JVET-AL0337 that the two complexity points of ~0.2x and ~5x VTM encoder time should be considered for evaluating CfE responses. This contribution studies the feasibility of achieving these points using ECM and VTM software with modified configurations parameters. It is concluded that the high complexity point (~5x EncT) can be easily achieved with ECM software with significant coding gain compared to VTM. However, ECM software within the lowest complexity point (~0.2x EncT) shows bad performance compared to VTM software. It is asserted that ECM requires high amount of code cleaning/simplification to meet VTM performance.

It is shown that ECM can still show reasonable performance at 2xVTM encoder run time, and gets similar performance at 1xVTM by switching off more tools. However, it gets worse at lower encoder run time.

It was commented that this is not surprising. ECM emerged from VTM, and was not targeting a faster encoding than VTM. Theoretically, by configuring it like 0.2xVTM, it should not become worse, but this may no longer be configurable with current software.

Proponents are asked to help AHG7 in fixing bugs when finding problems in disabling tools.