JVET-AM0069 Non-EE2: On Sign Prediction [Y. Zhang, L. Xu, Y. Yu, J. Gan, H. Yu, D. Wang (OPPO)]

This contribution proposes to include non-sign prediction coefficients in the calculation of hypotheses corresponding to the last two sign prediction coefficients. The modified hypotheses are used to calculate the boundary continuity for sign prediction. The method was implemented on top of ECM-17.0 and experimental results following CTC are summarized as follows:

AI: -0.01% (Y) -0.01% (U) 0.01% (V) 101.0% (EncT) 100.3% (DecT)

RA: -0.02% (Y) 0.02% (U) 0.06% (V) xx.0% (EncT) xx.0% (DecT)

LD: -0.09% (Y) 0.32% (U) 0.26% (V) 100.5% (EncT) 100.2% (DecT)

The number of inverse transforms needed for sign prediction is the same as ECM-17, which is 1.

It was commented that small gains were achieved in almost all classes and configurations. The added complexity seems to be manageable. Performance vs runtime increase is acceptable, except in AI case.

It was agreed to investigate this in an EE. Further optimization to reduce encoder runtime is requested.