For all three categories in the CE, it was indicated that the software was fully checked to ensure that it was doing what was intended.
For all of the subtests in Category A, the current HM design performed as well or better than any of the others, and the alternatives were not substantially simpler.
Category B considers a clipping range for minimizing motion vector storage precision for the temporal MV, either with variable MV precision or fixed MV precision, and also a (JCTVC-E221) proposed two modes of reducing the storage of MVs for different reference indices.
The benefit of the adaptive precision for temporal MV prediction does not seem adequate to justify the need to manipulate the data.
For the clipping ranges proposed for temporal MV prediction, the complexity benefit for imposing a more limited range did not seem to be sufficiently large to motivate changing the current design to limit the range.
The reference scaling mode from JCTVC-E221 / JCTVC-F278 did not seem to provide a sufficient degree of simplification to warrant changing the current design (with some loss of coding efficiency). This could be discussed again if some other motivation arises (e.g. from CE9 consideration).
For Category C, the test results seem to indicate that the factor-of-16 reduction in motion vector storage is sufficiently precise; only 0.1% bit rate savings can be obtained by disabling the motion data storage reduction completely.
Dostları ilə paylaş: | 
