Graphics Reference
In-Depth Information
a
b
temporal sub-layer 1
temporal sub-layer 1
B
2
B
4
P
1
P
3
I
0
P
1
P
3
I
0
P
2
P
4
temporal sub-layer 0
temporal sub-layer 0
Fig. 2.3
Temporal sub-layer examples
The following six bits contains a layer identifier that indicates what layer the
NAL unit belongs to, intended for use in future scalable and layered extensions.
Although the first version of HEVC, which was published in June 2013, supports
temporal scalability, it does not include any other scalable or layered coding so the
layer identifier (layer ID) is always set to '000000' for all NAL units in the first
version. In later versions of HEVC, the layer ID is expected to be used to identify
what spatial scalable layer, quality scalable layer, or scalable multiview layer the
NAL belongs to. These later versions are “layered extensions” to the first version
of HEVC and designed to be backwards compatible to the first version [
10
]. This is
achieved by enforcing that all NAL units of the lowest layer (also known as the base
layer) in any extension bitstream must have the layer ID set to '000000', and that
this lowest layer must be decodable by legacy HEVC decoders that only support the
first version of HEVC. For this reason, version one decoders discard all NAL units
for which the layer ID is not equal to '000000'. This will filter out all layers except
the base layer which can then be correctly decoded.
The last three bits of the NAL unit header contains the temporal identifier of
the NAL unit, to represent seven possible values, with one value forbidden. Each
access unit in HEVC belongs to one temporal sub-layer, as indicated by the temporal
ID. Since each access unit belongs to one temporal sub-layer, all VCL NAL units
belonging to the same access unit must have the same temporal ID signaled in their
NAL unit headers.
Figure
2.3
shows two different example referencing structures for pictures in a
coded video sequence, both with two temporal sub-layers corresponding to temporal
ID values of 0 and 1. The slice type is indicated in the figure using I, P, and B, and
the arrows show how the pictures reference other pictures. For example, picture B
2
in Fig.
2.3
a is a picture using bi-prediction that references pictures I
0
and P
1
,see
Sect.
2.4.4
for more details on prediction types.
Two very important concepts in order to understand HEVC referencing structures
are the concepts of decoding order and output order. Decoding order is the order in
which the pictures are decoded. This is the same order as pictures are included in
the bitstream and is typically the same order as the pictures are encoded, and is thus
also sometimes referred to as bitstream order. There are media transport protocols
that allow reordering of coded pictures in transmission, but then the coded pictures
