JVET-AD0128 AHG12: modified CTC proposal for low-delay configurations [F. Le Léannec, F. Galpin, K. Naser, T. Poirier, E. François (InterDigital)]

This contribution proposes to modify EE2 common test conditions, to significantly reduce the worst-case encoding runtime for LDB/LDP, together with slightly improved overall BD-rate results.

The CTC proposal consists in modifying encoder configuration as follows. The maximum MTT hierarchy depth is set to 4 in the I slice. For QP 22 of HD sequences, it is decreased from 3 to 2 for slices with temporal depth higher or equal to 2.

In LDB, the proposed modified CTC reportedly leads to 66% encoding time for the worst case (BQTerrace QP 22), 97% average encoding time relative to ECM-8.0 CTC, with -0.22% luma coding gain.

In LDP, the proposed modified CTC reportedly leads to 73% encoding time for the worst case, 100% average encoding time relative to ECM-8.0 CTC, with -0.16% luma coding gain.

The cross checker confirmed the performance results and the runtime reduction.

The proposed changes can be applied to the anchors (VTM11-ECM8) as well as the ECM itself.

It was commented that the encoding runtime impact seems to be inconsistent, where class B runtime was reduced significantly but class F runtime actually increased. This level of variation caused by the proposed change is considered undesirable.

It was commented that the proposed change to class F is applied on only one sequence, and that this type of sequence-dependent treatment is currently not done in ECM CTC. It was commented that sequence-dependent treatment could potentially result in skewed performance results on these “special” sequences in the future. This is considered undesirable.

It was commented that the proposed change would result in some increase in I slice quality but some decrease in B slice quality.

No action can be taken at this time. Further study.