Back to Search Document details
18th Meeting: Teleconference, April 2020 2020-04-03 08:12
AHG8/AHG9: On the derivation of PictureOutputFlag
Abstract
The current text for the derivation of the variable PictureOutputFlag normatively specifies a specific picture output behavior for an AU when the picture of the only output layer is not present (due to e.g. loss or layer down-switching). However, that piece of specification assertedly has multiple issues.
JVET-R0067 AHG8/AHG9: On the derivation of PictureOutputFlag [Y.-K. Wang (Bytedance), M. M. Hannuksela (Nokia)]

The current text for the derivation of the variable PictureOutputFlag normatively specifies a specific picture output behavior for an AU when the picture of the only output layer is not present (due to e.g. loss or layer down-switching). However, that piece of the specification assertedly has multiple issues.

This contribution proposes some changes to address the asserted issues, by either keeping the picture output behavior normatively specified but changed with the asserted issues fixed, or only describing it in a NOTE while in the specified normative picture output behavior the value of PictureOutputFlag for a current picture is set equal to 0 whenever the current picture does not belong to an output layer.

The authors initially suggested a preference for the normative output approach. Some others suggested that the decoder should be given some discretion if it encounters a “strange” situation, such as having a single target output layer and encountering an AU in which that layer is missing.

It was commented that the output behaviour in JVET-R0274 for ols_mode_idc modes 0 and 1 should be written to clarify that it is specifying behaviour at the AU level and that the picture output flag should also be considered. If that is clarified, the proposed normative in JVET-R0067 is the same as proposed in JVET-R0274.

Contribution JVET-R0123 was noted to be related, and essentially the same as the non-normative approach proposed here.

It was commented that mode 0 was originally intended for a use case where the encoder would intend that the highest received layer is the one that is output; otherwise a different mode would be used. However, this had not been everyone’s understanding of what the mode meant (where another interpretation is to just have a signalling shortcut).

It was suggested to define mode 0 (and mode 1) behaviour normatively but allow discretion for mode 2.

It was commented that we should not specify a “normative error concealment”.

Decision (bug fix / cleanup): Adopt the non-normative approach.

References:
JVET-Q2001
Decisions
adopted
Decision (bug fix / cleanup): Adopt the non-normative approach.
Citation