Block Structures and Parallelism Features in HEVC - High Efficiency Video Coding (HEVC) - page 68

Tabl e 3. 2 Summary of the test sequences for the two considered application scenarios that are
used in the comparisons. All sequences are given in the 4:2:0 chroma sampling format with a bit
depth of 8 bit per sample and have a duration of 10 s. The resolution is specified in luma samples
Entertainment applications

Interactive applications

Sequence name Resolution Frame rate Sequence name Resolution Frame rate

Kimono 1920 1080 24 Hz Four People 1280 720 60 Hz

Park Scene 1920 1080 24 Hz Johnny 1280 720 60 Hz

Cactus 1920 1080 50 Hz Kristen & Sara 1280 720 60 Hz

BQ Terrace 1920 1080 60 Hz Vidyo 1 1280 720 60 Hz

Basketball Drive 1920 1080 50 Hz Vidyo 2 1280 720 60 Hz

Vidyo 3 1280 720 60 Hz

The hierarchical prediction structure chosen for the entertainment application
scenario provides a very high coding efficiency, but it is associated with a structural
delay of eight pictures. For interactive video applications, such a high delay is not
acceptable. Here, we used a dyadic low-delay hierarchical prediction structure with
groups of four pictures [ 36 ] and four active reference pictures. For this coding
structure, all pictures are coded in display order, but the QP is varied depending on
the hierarchy level. Similarly as for the coding structure chosen for the entertainment
scenario, the QP is increased by 1 from one hierarchy level to the next. Except the
first picture of a video sequence, which is coded as I picture, all pictures are coded
as B pictures. The QP for the I picture is decreased by 1 relative to the QP for the B
pictures of the lowest hierarchy level. Furthermore, all pictures are coded as a single
slice. For the application scenario of interactive video application, we selected six
sequences with typical video conferencing content. All of these sequences have a
resolution of 1280 720 luma samples.
The chosen test sequences for both scenarios are summarized in Table 3.2 .They
represent a subset of the sequences that have been used by the standardization group
during the development of HEVC. Furthermore, the configurations for both selected
application scenarios are consistent with the test conditions [ 19 ] recommended by
the standardization group. All coding experiments were performed with the HEVC
reference software, version HM10.1 [ 20 ]. The encoder employs the Lagrangian
bit allocation technique described in [ 31 ]. For evaluating the impact of particular
aspects of the HEVC block structures and comparing it to the block structures
supported in older standards, selected block sizes for prediction and transform
coding were disabled in some of the simulations. This was partly done by specifying
corresponding parameters, such as the CTU size, the minimum CU size, or the
minimum and maximum transform size, in the encoder configuration files. Other
features were disabled by modifying the corresponding source code. In order to
compare the HEVC design to that of H.264 j MPEG-4 AVC, we additionally enabled
4 4 PUs in some experimental settings by slightly modifying the HEVC syntax.
The efficiency of different configurations was compared by encoding each of
the test sequences for an application scenario at ten different quantization settings.
Therefore, the QP for I pictures was varied from 20 to 38, inclusive, in steps of 2.