Organisation internationale de normalisation

3AVC base specification, errata, and related topics

The latest prior errata reporting status was provided in JVT-Z210.

3.1.1.1.1JVT-AB010 / VCEG-AI07 (Errata) [HHI] ITU-T AAP Last Call comments on H.264(2007)/Cor.1

This contribution contains various errata comments that were submitted as ITU-T AAP Last Call comments. Notes on particular topics in the contribution are provided below.

Regarding MMCO = 3, just delete "short-term" and add "(equal to 3 or 6)". In the following NOTE that contains "shall"s, change the "shall"s to "must"s and pluralize "field" at the end of the NOTE, and rephrase the extra NOTE as reviewed.
SPS activation for SVC – the SPS_id in the slice header can be different than the one in a BP SEI. The issue relates to the MGS feature. The precision of the position calculation depends on the level_idc, which may be different for different MGS layers. Two suggested approaches

• 
  use the level_idc of the layer with quality_id = 0 to determine the position calculation precision (decouple the active SPS for HRD purposes from the active SPS for position calculation purposes)
  
• 
  make the precision depend on frame size (width or height greater than 2048 or number of macroblocks greater than what's in 1920x1088 = 8160 Mbs) rather than level_idc.
  

Pick the first of those two options.
Active layer SPS should be activated by BP SEI – not just VCL NAL units. Agreed – seems obvious.
Removal of redundant constraint – yes – editorial.
Restriction for intermediate layers – yes – obvious.
Intra prediction constrained_intra_pred_flag and I_PCM – yes – obvious.
Clarification of overscan – minor reword – agreed.
4:2:2 and 4:4:4 – editorial/hypothetical – agreed.
Number of MVs per 2MB and bipred – 8x8 inference – touchy, but yes – agreed.
Motion constrained slice group SEI needs num_slice_groups_minus1 for parsing – add a NOTE suggesting to store the RBSP until the active PPS is known, then parse – agreed.
Two rewriting issues – no action needed on either item.
End of sequence NAL unit in SVC context – all layers must have IDR after that – agreed.
Persistence of SVC SEI messages – needs further study by editors. E.g., scalability information SEI persists forever – say that it ends if there's IDR at all layers.
Discretion given to editors for determining aspects not noted and finalizing according to their best understanding.

3.1.1.1.2JVT-AB011 / VCEG-AI08 (Errata) [Microsoft] ITU-T AAP Last Call comments on H.264(2007)/Cor.1

This contribution contains various errata comments that were submitted as ITU-T AAP Last Call comments. Notes on particular topics in the contribution are provided below.

Comment 5(c): Seems to only be an issue when using Tools->Options->View-All, so it may not be necessary to fix.
Comment 5(k): Specify that the decoder shall ignore the content of the SEI message in those cases.
Comment 5(m): Re-examine aspects relating to "overscan", including in particular "do neither use".
Comment 5(n).i: We may have too many cases like that to try to fix them all, and in this case you can include the opening parenthesis in the search.
Comment 5(n).iii: May have already been fixed.
Comment 5(o): Maybe that's OK as-is.
Further details to be addressed in editing work.

3.1.1.1.3JVT-AB012 / VCEG-AI09 (Errata) [Sony] ITU-T AAP Last Call comments on H.264(2007)/Cor.1

See discussion below regarding JVT-AB013.

3.1.1.1.4JVT-AB013 / M15558 (Errata) [JNB of WG11] WG 11 NB comments on ISO/IEC 14496-10:200X/DCOR1

Discussed together with JVT-AB012.
These contributions advocate not to delete certain problematic level limits that had been specified to apply to (only) the first access unit of the bitstream. A.3.3.a and G.10.2.2.a: Number of slices in access unit 0. A.3.1.c, A.3.3.h, G.2.2.h: Number of bytes for access unit 0.
Remark: The problem of excess load possibility was sort of already there, since timing information could be stretched by an encoder to cause "hostile" bitstream characteristics.
JVT Disposition: These limits will not be entirely removed, however, they will not be made overly tight either – exact details are left to be worked out by the editors and interested parties and editors are to put their best effort outcome into the draft.

3.1.1.1.5JVT-AB021 (Prop 2.2/3.1) [K. Kazui, A. Nakagawa, S. L. Bhavani, S. Kobayashi (Fujitsu)] Modification of POC type 1 specification

At the previous meeting in Geneva, JVT-AA022 it was asserted that the current specification of AVC does not always guarantee correct recovery of relative POC value when POC type 1 is used and a decoder starts the decoding process from a recovery point rather than an IDR picture. JVT-AA022 proposed three candidate actions to address the issue.
Remark: Because the semantics of the recovery point SEI message include a "promise" made to the decoder that the decoder can recover the correct pictures if it operates in a certain manner, therefore an encoder is not allowed to put the RP SEI message into the bitstream if such a use would create a non-recovery behavior. Agreed.
Since no prescription has been established in the text, the decoder should, in principle, able to initially guess any value of POC within the 32 bit range for the current picture and for any reference to a non-existing picture, and allocate a greater than 32 bit dynamic range to its POC estimate for subsequent delta accumulations and then everything must work properly once the recovery_frame_cnt has gone by. (It is also hypothetically possible to keep the value within a 32 bit range by adjusting the value of POC on every picture in the DPB by an offset whenever a value would otherwise be created that would violate the 32 bit range.)
It is the encoder's responsibility to ensure that no matter what value of POC the decoder initially guesses, once the recovery_frame_cnt has gone by, everything must work.
Add a NOTE to the spec to clarify this understanding.
The encoder must not (i.e. shall not) send a RP SEI message that would put the decoder into an undefined output state that persists beyond recovery_frame_cnt.
The RP SEI establishes a promise. The promise must be fulfilled.
Agreed.

3.1.1.1.6JVT-AB036 (Errata 2.2.1/3.1) [Y.-K. Wang (Nokia)] SVC errata items

Some issues reported here were covered in the review of JVT-AB010.

Editor asked to study/resolve use of "RBSP" terminology relating to parameter sets. Agreed.
Editor to ensure proper specification of constraint on "active SPS" as well as "active layer SPS". Agreed.
Profile definitions – decoder conformance requirement for decoding ordinary AVC bitstreams should specify decoding of all ordinary compatible AVC bitstreams in a manner equivalent to what applies in Annex A. Agreed.
On restrictions to scalable nesting SEI message clarify that the "less than any of the particular values…" refers to the values of ( sei_dependency_id[i] << 4 ) + sei_quality_id[i], not the value of i. Agreed (insert commas around "for all present i" or otherwise restructure logic to clarify).