Graphics Reference
In-Depth Information
Tabl e 5. 3
Average bit rate
losses for different merge
estimation regions in terms of
BD-rate using HM5.0
reference software
Merge estimation region
8
8 6
16
32
32
64
64
RA-HE
0.1 %
0.6 %
1.6 %
2.7 %
LB-HE
0.2 %
0.7 %
2.0 %
3.4 %
the merge candidate list. In that case, the merge list of PU5 consists only of the
temporal candidate (if available) and zero MV candidates.
In order to enable an encoder to trade-off parallelism and coding effi-
ciency, the parallel merge estimation level is adaptive and signaled as
log2_parallel_merge_level_minus2
in the picture parameter set. The
following MER sizes are allowed: 4
4 (no parallel merge estimation possible),
8
8, 16
16, 32
32 and 64
64. A higher degree of parallelization, enabled by
a larger MER, excludes more potential candidates from the merge candidate list.
That, on the other hand, decreases the coding efficiency. The coding efficiency
losses in terms of BD-rate [
2
] for different MER sizes are reported in [
28
]and
summarized in Table
5.3
. Results are generated using HM5.0 [
13
] with random
access high efficiency (RA-HE) and low delay B high efficiency (LB-HE) coding
configurations as described in [
3
].
When the merge estimation region is larger than a 4
4 block, another modifica-
tion of the merge list construction to increase the throughput kicks in. For a CU with
an 8
8 luma CB, only a single merge candidate list is used for all PUs inside that
CU.
5.2.3
Motion Data Storage Reduction
The usage of the TMVP, in AMVP as well as in the merge mode, requires the
storage of the motion data (including motion vectors, reference indices and coding
modes) in co-located reference pictures. Considering the granularity of motion
representation, the memory size needed for storing motion data could be significant.
HEVC employs
motion data storage reduction
(MDSR) to reduce the size of the
motion data buffer and the associated memory access bandwidth by sub-sampling
motion data in the reference pictures [
20
]. While H.264/AVC is storing these
information on a 4
4 block basis, HEVC uses a 16
16 block where, in case of
sub-sampling a 4
4 grid, the information of the top-left 4
4 block is stored. Due
to this sub-sampling, MDSR impacts on the quality of the temporal prediction.
Furthermore, there is a tight correlation between the position of the MV used in
the co-located picture, and the position of the MV stored by MDSR.
During the standardization process of HEVC, the impact of the sub-sampling
scheme as well as the interaction with the TMVP was investigated in a core
experiment on MDSR [
15
]. It turned out that storing the motion data of the
top left block inside the 16
16 area together with the bottom right and center
