Of itu-t sg16 wp3 and iso/iec jtc1/SC29/WG11

JCTVC-M0168 On random access point for HEVC extension [B. Choi, Y. Cho, M. W. Park, J. Y. Lee, J. Yoon, H. Wey, C. Kim (Samsung)]

Proposes to allow BLA pictures to not be cross-layer aligned. Relates closely to M0206 / M0207 / M0266. This aspect for further study. See notes on those contributions.

A second aspect proposed that an IRAP NAL unit of each layer with NoRaslOutputFlag equal to 1 may activate a new SPS for the corresponding layer. Decision: This was agreed.

The contribution discussed handling of RASL pictures in an enhancement layer, and the proposal seemed to request something that was already intended in the design – which can be clarified as necessary – i.e. that RASL pictures may occur in an enhancement layer.

Issues relating to higher layer pictures referencing RASL pictures in a lower level were discussed in the contribution. Participants suggested to try to establish constraints on what an encoder can do, in order to prevent pathological cases (rather than having the decoder react to certain conditions).

The proposal suggests to enable some IDR / CRA / BLA properties to be indicated in the AUD.

It was noted that the current specification does not specify how the decoder should react to an AUD with pic_type outside the range of 0..2. It was agreed that this was an oversight, since having a 3-bit syntax element with only two bits actually used seems to be a clear indication that the intent was to have some values reserved for future use and ignored if present. Corrective action (e.g. corrigendum if necessary) was agreed to be desirable to fix this editorial oversight.

Regarding the particular properties which we could indicate in the reserved pic_type values, further study was suggested. It also remained for further study or later resolution during the meeting to consider whether we will introduce new slice types that could indicate certain properties, and what cross-layer alignment constraints to impose on different IRAP types.

It was also discussed whether having the 4th bit not equal to 1 should also be considered a reserved case. It was agreed that this should just be considered non-conforming rather than reserved.

The allowed layer ID value for the AUD was discussed, and it was agreed that the value should correspond to the lowest VCL NAL unit layer ID in the AU.