JVET-AM0072 Non-EE2: Diversity reordering for ARMC pairwise candidates [C. Wang, Y. Yu, L. Zhang, Z. Xie, H. Yu, D. Wang (OPPO)]

This contribution proposes a modification to the process of Adaptive Reordering Merge Candidates list (ARMC). It is proposed to apply diversity reordering after the first reordering process and before deriving the pairwise-average merge candidates, aiming to establish a more diverse list of merge candidates for the pairwise-average merge. The test results on top of ECM-17.0 are summarized as follows:

RA: -0.01%/0.01%/0.00%; EncT X% DecT X%；

LDB: -0.03%/0.21%/0.16%; EncT X% DecT X%；

Runtime roughly constant for RA, but some increase in LB for decoding runtime (less in cross-check, may be unreliable).

Some additional checks necessary at decoder, and gain is not significant (also not consistent across classes). Further study was recommended.