International organisation for standardisation

6.10.7Decoded picture buffering

6.10.7.1.1.1.1.1.1JCTVC-F381 Clean decoding refresh definition and decoding process [T. K. Tan, A. Fujibayashi (NTT Docomo)]

This contribution proposed a change to the decoding process relating to "clean decoder refresh".

It proposes to define a "CDR reset picture".

The special decoding process aspect of the proposal did not seem to be strictly necessary. See additional notes below relating to JCTVC-F464 and JCTVC-F759.

6.10.7.1.1.1.1.1.2JCTVC-F464 On CDR picture [Y.-K. Wang (Huawei), M. M. Hannuksela (Nokia), Y. Chen (Qualcomm)]

This document proposed some changes to clarify the definition of clean decoding refresh (CDR) picture, including a change of term to “clean random access” (CRA), as well as other editorial changes.

It was suggested that this is only an editorial refinement of the definition previously agreed.

A problem was pointed out in the proposed wording during the discussion (again, a matter of clarification of the intent rather than dispute over the intent).

An improved definition was suggested to be conceptually structured as follows:

• 
  “XYZ” pictures = pictures that have both decoding order and output order greater than or equal to that of the CDR picture
  
• 
  XYZ pictures shall not refer to non-XYZ pictures
  

(Some clarification of "refer to" was suggested to potentially be needed.)

Decision: This was agreed.

Further study regarding whether to make the decoder ref pic list construction and/or ref pic marking processes depend on this.

A BoG (coordinated by Y.-K. Wang) was asked to further refine the definition consistent with this and was asked to think about the potential decoding process dependency. The concept of "lagging pictures" was also later discussed and fixed in the development of this definition. See notes relating to the BoG report JCTVC-F759.

6.10.7.1.1.1.1.1.3JCTVC-F604 Detection of CDR for random access [Y. Park, K. P. Choi, C. Kim (Samsung)]

The clean decoding refresh (CDR) picture as a new type of random access points was introduced to HEVC and the discussion related to CDR was proposed in JCTVC-E400. In this proposal, a specific method is proposed to detect the current status of whether the decoder is performing random access into a position in a different CVS or is conducting normal play when a CDR picture is decoded.

The contribution proposed inserting a discontinuity counter into the slice header and SPS.

It was remarked that this counter may create some problems, e.g., regarding very long bitstreams or splicing.

It was remarked that a system may have some other way to indicate this to the decoder. However, the proponent indicated that it may be desirable to not need to depend on such an external signal.

For further study.

6.10.7.1.1.1.1.1.4JCTVC-F493 Absolute signalling of reference pictures [R. Sjöberg, J. Samuelsson (Ericsson)]

This document contains a proposal regarding the reference picture processes in HEVC. It is proposed that pictures that should be available for reference are described in an absolute manner in the slice header of a picture instead of in a relative manner (through MMCO and sliding window process). It is further proposed that buffer operations (picture marking) is performed directly after parsing of the first slice header of a picture instead of after the picture has been decoded.

The document claims that the proposed changes improves error resilience, simplifies temporal scalability and enables decoding at a minimum number of decoded picture buffer slots.

The proposal removes frame_num, instead uses only POC, includes a buffer description in the slice header (or a reference to a table of such descriptions sent in the PPS). Removes MMCO commands.

The proposal uses temporal_id to identify relevant pictures.

Question: Can a lower layer picture mark an upper layer picture as unused for reference?

The encoder includes all of the higher temporal layer pictures in the buffer description.

The proposal has a variation in which buffer descriptions are moved into the PPS.

The proposal does not support pictures held in the buffer for a long time (beyond max POC wrap) before output.

The group considered the proposal very interesting.

It was suggested to put the buffer description in a slice parameter set / adaptation parameter set.

Do we have syntax and semantics and decoding process? Partial, not complete.

It was suggested to establish and AHG to examine, analyze, identify any needs for clarification, and produce a (single) clarified text specification (not necessarily including all features).

It was asked whether we should make a source code branch for this, and agreed that yes, we should do that.

6.10.7.1.1.1.1.1.5JCTVC-F681 Cross-check of Ericsson proposal JCTVC-F493 by Sony [Y. Morigami, K. Sato (Sony)] [late reg. 07-06, upload 07-12]

Did not study the source code – just ran some simulations.

6.10.7.1.1.1.1.1.6JCTVC-F460 Getting rid of non-existing pictures [Q. Shen, Y.-K. Wang (Huawei)]

This document proposed to remove the processes for generating and handling of “non-existing” pictures in the HEVC decoding process. The proposed changes include a change to the frame_num semantics, some alternative changes to the reference picture list construction process, and some changes to the reference picture marking process. Assuming that in HEVC the hypothetical reference decoder (HRD) process is specified basically the same as AVC; then those HRD texts mentioned “non-existing” pictures are also proposed to be changed. An example of such changes to the HRD process is also provided, for information.

The proposal suggests that now that we have temporal_id, we may not need the "non-existing" pictures concept.

It is primarily a proposal to simplify the specification (rather than the actual implementation).

This proposal would be obsoleted if the JCTVC-F493 proposal is adopted. Further study was encouraged in case something closer to JCTVC-F493 is not adopted.

6.10.7.1.1.1.1.1.7JCTVC-F461 Reference picture loss/error detection [Y.-K. Wang, Q. Shen (Huawei)]

This document proposes to conditionally mandate the use of reference picture list modification (RPLM) commands for construction of reference picture lists, to enable decoders detect reference picture losses/errors when gaps_in_frame_num_value_allowed_flag is equal to 1, for applications in error-prone environments. The proponent indicate that if the proposal of unconditionally mandating RPLM commands in Section 6.2 of JCTVC-F460 is agreed, then this proposal can be ignored.

This proposal would be obsoleted if the JCTVC-F493 proposal is adopted. Further study was encouraged in case something closer to JCTVC-F493 is not adopted.

6.10.7.1.1.1.1.1.8JCTVC-F462 On reference picture marking [Y.-K. Wang (Huawei)]

This document proposes some changes to the reference picture marking process. Specifically, it is proposed that, at any access unit, only reference pictures with an identical or greater value of temporal_id may be marked as “unused for reference”. When “non-existing” pictures are involved, it is proposed to be assumed that “non-existing” pictures have the greatest possible value of temporal_id, i.e., 7.

The first aspect seems like just a clarification of the existing intent. Agreed.

The second aspect we are less sure about, but it would be obsoleted if the JCTVC-F493 proposal is adopted. Further study was encouraged in case something closer to JCTVC-F493 is not adopted.

6.10.7.1.1.1.1.1.9JCTVC-F546 Sliding Window Improvement for Temporal Scalability [Y. Chen, P. Chen, M. Karczewicz]

Duplicate content to part of JCTVC-F462. See notes on JCTVC-F462.