International organisation for standardisation organisation internationale de normalisation

5.13.2.2Other

The proposals in this category were reviewed with the assistance of summary information prepared by Y.-K. Wang, and recorded as follows.

	Method summary	Comments
H0071	Merging APS into PPS + PPS partial update = No APS; Diff types of PPS params: existing, “non-adaptive” PPS params, plus existing different types of APS params	Decreases the possibility of out-of-band PPS transmission in session negotiation or announcement Currently, the possibility of out-of-band APS transmission is not high, and not possible when some APS data change very frequently and the total number of variants is large Moving APS data into PPS would significantly decrease the possibility for out-of-band PPS transmission.
H0132	Reported that APS loss may cause serious artifacts. Proposes to enable repeating of APS parameters in SH (either use from APS or use from SH)	Why repeating in slice header is better than repeating the APS itself?
H0376	More params from SH to APS; Fragmentation of ALF and SAO for low delay	Low-delay issue can be solved by allowing different slices referring to different APSs Would fragmentation of ALF and SAO data be useful for any other purposes than low-delay support?
H0507	Allowing referencing different APSs in different slices of a picture
H0512	APS fragmentation for QM for MTU size matching	Not a generic APS fragmentation mechanism, only provides fragmentation of QM data On the other hand, existing APS syntax supports to split different types of APS data into different APS NAL units

5.13.2.2.1.1.1.1.1JCTVC-H0071 On the need for an Adaptation Parameter Set [S. Wenger, J. Boyce, Won Kap Jang (Vidyo)]

With the (in the authors' opinion) possible, even likely, inclusion of some form of partial parameter set update into HEVC, it was suggested that the need for an APS may not exist anymore. This document analyses the implications of moving APS parameters into the PPS and relying on a partial update mechanism for picture-adaptivity. The size of the slice header could reportedly be reduced with little or no increase in the size of the parameter sets, and the number of NAL units required for coding a picture could be reduced, saving bits for NAL unit headers, start codes or other segmentation information. However, currently in the draft are only very few table entries for APSs and a lot of entries for PPSs. If the APS content were merged with the APS content, it was asserted that either the number of table entries for the PPS needs to be reduced, or the memory requirements in the decoder would increase.

For further study.

5.13.2.2.1.1.1.1.2JCTVC-H0132 On Adaptation Parameter Signalling [Y. Ye, E. S. Ryu (InterDigital)]

In this proposal the error propagation effect of losing SAO and ALF parameters of the APS was studied by analysing the case of losing only the first APS. Quality degradation in terms of both PSNR and visual quality was analysed. It was reported that error propagation due to the loss of SAO and ALF parameters is content dependent. In terms of PSNR degradation, the maximum loss is 2 dB for Y, 4 dB for U, and 5 dB for V. For all three components, the minimum PSNR loss is reportedly close to 0 dB. In terms of subjective quality, visual degradation ranges from unnoticeable for some test sequences to significant visual artefacts for others. It is reported that for those sequences where visual artefacts were observed, at which frame and for how long the artefacts appear is also content dependent; in other words, some artefacts appear early in the sequence while others appear later in the sequence, and some artefacts appear for just a few frames while others propagate throughout many frames. Based on these observations, a method to improve error resilience of adaptation parameters carried in APS was proposed.

This contribution tested HE LB encoding. The loss effects reportedly did not correlate with QP value.

The most visible artefacts tended to be chroma shifts.

The contribution proposed a flag for the slice header which would indicate sending the APS information in the slice header rather than in the APS.

It was remarked that repetition of the APS outside of the slice header (as an APS NAL unit or hypothetically in some SEI message) would seem to be a (probably better) alternative way to get the desired loss resilience.

It was remarked that LCU-based SAO and ALF may reduce the dependence on the APS.

5.13.2.2.1.1.1.1.3JCTVC-H0376 Non-CE8: Low-delay support for APS [S. Esenlik, M. Narroschke, V. Drugeon, T. Wedi (Panasonic)]

Several proposals related to LCU-based Adaptive Loop Filter (ALF) and Sample Adaptive Offset (SAO) have been submitted. LCU-based syntax for ALF and SAO provides the possibility of low encoding delay for the two in-loop filters. This contribution proposed a mechanism to allow usage of APS in low delay scenarios in connection with LCU-based ALF and SAO approaches. The proposal uses the APS to carry the filtering information that is shared by multiple slices in a frame. Moreover a fragmentation scheme was proposed. As opposed to signalling all of the filtering parameters inside the slice, the proposed low-delay APS fragmentation scheme is asserted to not suffer from reduction in the ALF and SAO coding gain in the case where a frame is partitioned into multiple slices.

H0507 was suggested to have a similar spirit and to perhaps be a more generic approach.

5.13.2.2.1.1.1.1.4JCTVC-H0507 Referencing different APSs in the same picture [S. Wenger (Vidyo), Y. -K. Wang, I. S. Chong (Qualcomm)]

This contribution proposed the option to activate more than one adaptation parameter set (APS) per coded picture. As a result, at least some of the coding tools having parameters that are carried by the APS (such as loop and deblocking filters, SAO, and quantization matrixes) can be adapted to different content in a picture, by appropriately placing slice headers into the picture referencing different APSs (or changing the APS content while re-using the same APS ID).

The H0376 and H0507 proposals are interesting, but seem to depend on other decisions that were yet to be made. Further study was encouraged.

5.13.2.2.1.1.1.1.5JCTVC-H0512 APS fragmentation [Y. Chen, Y. -K. Wang, R. Joshi, M. Karczewicz (Qualcomm)]

This contribution ws a follow-up proposal of JCTVC-G658. In JCTVC-G658, it was proposed that the quantization matrices be signalled in the adaptation parameter set (APS). To solve the problem that an APS can be larger than the maximum transmission unit (MTU) size, it was proposed that the APS could be fragmented into multiple NAL units, each of which can be independently parsed and applied. The problem was reportedly confirmed to be a valid problem at the previous meeting. In this proposal, the same idea was proposed again, based on the updated syntax element support for the quantization matrix signalling in the APS, such that an APS, if it contains information for quantization matrix, may contain only a subset of the total number of quantization matrices.

It was commented that it appeared that the suggested functionality seems achievable using some of the other proposed approaches.

5.13.2.2.1.1.1.1.6JCTVC-H0667 Implementation issues with frequent slices [C. Fogg, A. Wells (Ambarella)] [late 02-04]

Rather than limit the frequency of HEVC slices, a potentially viable scheme would be to restrict changes to subsets of slice header elements that are typically nearly constant in applications but have an asserted heavy impact upon implementation when changed. An analysis was included to identify those subsets. Based on the results presented, it was proposed that profiles and levels limit deblocking control to no more than four modifications per frame. Finally, it was suggested to move slice header element sets identified in the contribution as being fairly static into a slice parent-level syntax structures (e.g. APS or PPS) that have suitable frequency restrictions.

Deblocking loop filter control and, to a lesser extent, reference picture list modification were identified as relatively important.

It was also mentioned that, depending on design choices, SAO and ALF parameters may be an additional source of difficulty.