12.3.2Second part of CE9: Adaptation and modification of the set of predictors
JCTVC-D301 "3.2.c" proposes reordering the predictor to put the temporal predictor first. It was tested with merging off, and provided no reported benefit relative to merge enabled in the anchor.
A cross-checker JCTVC-D117 tried enabling merging in combination with this and reported a benefit. The combination reportedly provides about 0.2% benefit in the HE RA case (relative to merge enabled in the anchor). It was remarked that a previous similar test was done in Guangzhou with better results.
JCTVC-D164 is related.
JCTVC-D231 "3.2.d" proposed three simplifications: 1) removing the median predictor candidate, 2) disabling a pruning process of MV prediction candidates based on MV differences, 3) changing the method of choosing the MV predictors for the left and above predictors and reducing the number of candidate predictors. It was reported that the first of these modifications results in no loss in coding efficiency, the second results in 0.2% average loss, and the third results in 0.2% average gain.
JCTVC-D303 was suggested to be related – corresponding to the first topic investigated in 3.2.d.
JCTVC-D055 was suggested to be related to the third aspect.
Decision: These three simplifications (from JCTVC-D231) were adopted.
Dostları ilə paylaş: | 
