JVET-AM0125 AHG17: Analysis on encoding time and mode cost test for draft CfE sequences [Y. Tokumo, S. Hong, T. Ikai (Sharp)]

This contribution reports analysis on encoding time and mode cost test (the number of RDO search, RDO count) for draft CfE sequences. Experimental results show that runtime ratio of VTM high performance and ECM JVET-AL0245 are roughly 1.3x to 1.9x and 10.4x to 5.3x for R1 to R4, and encoding runtime and RDO count are correlated, and they vary greatly depending on sequences. For SDR_RA_HD, EncT per RDO count is 0.087, 0.082 msec/count in VTM 23.9 under default and high performance configurations respectively. On the other hand, it is 1.060 msec/count in ECM under JVET-AL0245 configuration, which is 13 times larger than that of VTM and it may reflect to 12x DecT in ECM to VTM.

v2 added/fixed experimental results and added the following proposal.

Considering ECM/VTM EncT ratio in CfE operation points are much higher than those of CTC. It is proposed to apply encoding times condition or plot to the highest rate point, R4.

One finding is that the runtime per RDO check is significantly higher in ECM, which may be due to the slower decoding process that needs to be conducted during RDO.

The number of RDO checks is lower in ECM than VTM, which according to the proponents’ analysis I likely due to the MTT depth reduction.

It was commented that RDO count in ECM may be incomplete, as the tracing functionality may not be implemented in all tools

It was commented that only analysing R4 in encoder complexity would not be appropriate. Further discussion is necessary on that, as the number of data is large, which could be done by analysing the CfE results.