SVC标准G.8.1.4.1中关于ref list construcion的一个问题

from：http://forum.byr.edu.cn/pc/pccon.php?id=1727&nid=77191&order=&tid=1329

————————问题————————
在标准G.8.1.4中关于reference list construction，可以看到以下的描述：

-       (G.8.1.4.1) When currDQId is equal to 0 and (slice_type % 5) is
equal to 1, the SVC decoding process for reference picture lists construction as specified in subclause G.8.2.3 is invoked.

-       (G.8.1.4.2) When CroppingChangeFlag is equal to 1 and (slice_type %
5) is less than 2, the SVC decoding process for reference picture lists construction as specified in subclause G.8.2.3 is invoked.

在这里reference list construction的用意是什么？为什么在G.8.1.4.1中只对currDQId = 0 的B slice生成参考队列呢？而在G.8.1.4.2中，又为什么当CroppingChangeFlag =1时生成队列呢？

————————mail list————————
Q：
Dear experts,

In G.8.1.4, there are some descriptions about the reference list
construction for the decoding of slices.

-   (G.8.1.4.1) When currDQId is equal to 0 and (slice_type % 5) is equal to
1, the SVC decoding process for reference picture lists construction as
specified in subclause G.8.2.3 is invoked.

-   (G.8.1.4.2) When CroppingChangeFlag is equal to 1 and (slice_type % 5)
is less than 2, the SVC decoding process for reference picture lists
construction as specified in subclause G.8.2.3 is invoked.

The later case is quite reasonable since ref lists are needed for both B and
P slices (slice_type % 5 is less than 2). But the former one is puzzling. I
think it means the ref list construction is invoked for the B slices
(slice_type % 5 is equal to 1) but skipped for P slices (slice_type % 5 is
equal to 0). Why there's no construction for list0? I wonder if it is
completed in other places. If so, could someone point it out for me?

Thanks for any information that you can provided.

Best regards,
Wenhao

A：
The reference picture list 0 is not needed for the part of the decoding
process specified in subclause G.8.1.4.1. And reference picture list 1 is
only needed for a B slice of the base layer (that's why it is only
constructed if currDQId is equal to 0 and slice_type specifies a B slices).
- The only reason why we need reference picture list 1 for this particular
case is that it may be required for deriving the motion data in B slices of
the base layer (for SVC enhancement layers the co-located motion data are
not used).

The actual motion compensated prediction is done at a later stage in the
decoding process: in the target layer representation decoding process
specified in G.8.1.3.3. And here, both reference lists (as far as
applicable) are derived (in step 3d).

Best regards,
Heiko

Q:
Thanks for the detailed explanation.

Here're some further questions. You said for SVC enhancement layers the
co-located motion data are not used. It means if an EL block uses temporal
direct method, the co-located motion data are from the BL, but not from the
EL, right? And is this statement always true for both the scenarios with
resolution change and without resolution change?

Thanks in advance.
Wenhao

A:

The temporal direct mode is not supported in an SVC bitstream (see G.7.4.3
and G.7.4.3.4 - both subclauses specify "The value of
direct_spatial_mv_pred_flag shall be equal to 1"). Furthermore, the spatial
direct mode has been modified for SVC enhancement layers as specified in
subclause G.8.4.1.2 (it doesn't require the analysis of co-located motion
data). As a result, co-located motion data are only needed for checking the
"special condition" for the spatial direct mode of the base layer (DQId =
0).

-Heiko

————————总结————————
1. G.8.1.4.1中的ref list1生成的目的是为了推出B direct方式下的mv预测值，因为此时必须要有list1的第一帧的具体mv、ref index信息才能反推mv的预测值。而其它途径的mv预测只需要知道BL层或者同层相邻block的mv，无需生成ref list。此外这里也说明，当without resolution change的时候，EL层的B direct的mv推导是利用BL的mv信息得到的。Schwarz的这句话For SVC enhancement layers the co-located motion data are not used表达的应该是这个意思。

1. 在SVC中，无论是BL还是EL，只有spatial direct而没有temporal direct。

2. G.8.1.4.1中的ref list1生成的唯一目的为了推出base layer的B spatial direct方式下的必要的co-located信息，因为此时必须要有list1的第一帧的motion information。对于enhancement layer，为什么不需要呢？需要注意的是SVC的enhancement layer的spatial direct为什么不需要生成ref list1，(G.7.4.3.4 - "The value of direct_spatial_mv_pred_flag shall be equal to 1")，是因为enhancement layer的spatial direct方法有所修改，不再需要co-located motion data（G.8.4.1.2）。

2. G.8.1.4.2中为什么当CroppingChangeFlag等于1的时候就要对P或者B生成ref list。原因是在ESS的mv推测过程中，需要时域的信息。具体可参见G.8.6.1.2中的变量tempRefIdxPredLX[x,y]。

3. MC所需要生成ref list是在G.8.1.3.3 target layer representation decoding process中。