Organisation internationale de normalisation
Non-normative: Encoder optimization, decoder speed improvement and cleanup, post filtering, loss concealment, rate control (10)
Yüklə 9,04 Mb.
|
Non-normative: Encoder optimization, decoder speed improvement and cleanup, post filtering, loss concealment, rate control (10)13.0.0.1.1.1.1.1.332JCTVC-S0067 Combination of several encoder improvements for SCM2.0 [G. Laroche, C. Gisquet, T. Poirier, P. Onno (Canon)] (Consideration of this topic was chaired by G. Sullivan on Sunday 10-19 a.m.) This contribution gathers several improvements related to IBC, palette coding mode and the adaptive residual colour transform. It is reported that these proposed encoder changes provide a BDR gain of 2% over SCM2.0 while decreasing the encoding runtime to 76% for Intra configuration. It is also asserted that the resulting combination offers a better reference for evaluating the complexity/gain trade-off of normative proposals over SCM2.0. There were three elements of the contribution:
The palette-related improvements for this proposal are identical to JCTVC-S0066. The other improvements are not related to palette (e.g. IBC-related). It was asked how "clean" are the code changes for items 1 & 3. It was remarked that knowing the separate impact of the changes would be desirable. However, the benefit seems substantial. Partial cross-check was available as S0261. Decision (S/W): Adopt, items 1 & 3. (The proponent also said they would help improve the text description of the test model algorithms.) See also notes on S0149. 13.0.0.1.1.1.1.1.333JCTVC-S0282 Crosscheck of JCTVC-S0067 [F. Zou (Qualcomm)] [late] (Has "the" problem in abstract phrasing.) 13.0.0.1.1.1.1.1.334JCTVC-S0149 Fast intra coding mode decision for screen content coding [Y. Ahn, X. Wu, W. Lim, J. Ma, D. Sim (KWU)] (Consideration of this topic was chaired by G. Sullivan on Sunday 10-19 p.m.) In this contribution, an early termination of intra coding mode decision is proposed to reduce an encoding complexity of the screen content coding (SCC) extensions. Comparing to HEVC version 1, the computational complexity of SCC extensions has been increased due to newly added intra coding mode as intra block copy (IBC). To reduce complexity from additional mode decision in encoder, adjustment of decision order for intra coding modes and early termination based on coding information of IBC 2Nx2N mode are proposed in this contribution. The proposed method reportedly achieved about 10% reduction in encoding time with BD-bitrate benefit of 0.3% compared with that of SCC extensions test model 2.0 encoder in all intra (AI) case. The decision order is modified to check IBC 2Nx2N before checking ordinary-intra 2Nx2N, and early termination checks are used. S0065 and S0067 are related. The technique discussed in this contribution is included in S0067. See notes on that contribution. 13.0.0.1.1.1.1.1.335JCTVC-S0089 Improvement for hash based inter search [B. Li, J. Xu (Microsoft)] (Consideration of this topic was chaired by J. Boyce on Sunday 10-19 p.m.) This document proposes non-normative improvement for hash based ME. With the proposed method, the encoding time is reduced by 7% for lossy RA coding and lossy LB coding, without significant impact on the coding efficiency. The coding impact was minimal, but there was loss of up to 0.3%, and gains up to 0.1%. The encoder speed gains were sequence dependent. In the current design, homogenous blocks are not added to the hash table. In the proposal, CU-aligned homogenous blocks are added. A participant questioned why to use CU-alignment position as a criterion. The proponent asserted that using CU-alignment was a way to minimize the additional memory required for the proposal. Decision (SW): Adopt into the test model SCM. Will also need to update test model document. 13.0.0.1.1.1.1.1.336JCTVC-S0236 Crosscheck of Improvement for Hash Based Inter Search (JCTVC-S0089) [W. Zhang, L. Xu, Y. Chiu (Intel)] [late]
(Consideration of this topic was chaired by J. Boyce on Sunday 10-19 p.m.) This document proposes a modified configuration to support temporal scalability when the random access coding structure is used. Compared with RA anchor, which does not support temporal scalability, enabling temporal scalability with the modified configuration brings 0.1% performance loss on average. When compared with the current configuration supporting temporal scalability in SCM-2.0, i.e., applying the referencing structure provided in encoder_randomaccess_main_4tids.cfg to encoder_randomaccess_main_scc.cfg, about 0.3% performance gain is achieved by the modified configuration. The contribution was proposing a change to common test conditions. The proposal reduces the penalty to use temporal scalability for RA. The proposed referencing hierarchy does not support temporal nesting. Decision (SW): Adopt the change to the configuration file for temporal scalability. Also fix the bug in the SCM that was identified. Further discussion in the JCT plenary was chaired by GJS on Thursday 10-23. In the discussion, it was asked whether the recorded adoption for "temporal scalability" was intended to express a modification of the ordinary random access operation for the HM. It was asked whether there would be any problem with such a modification and what would be the effect of the change on camera-view content. The contributor said they had tested this and had reported results in a previous contribution JCTVC-R0103, which used the same reference structure in the HM and tested the performance with class A through F sequences. The average loss was reportedly 0.1% for both RA_Main and RA_Main10. During the discussion, it was commented that with such an insignificant difference in coding performance, it might be good to apply the temporal scalability reference structure to all RA cases. Thus it was agreed that the HM random access configuration should use this form of temporal scalability. It was discussed whether the SCM RA common conditions should be changed to support temporal scalability. It was commented that SCM ordinarily tracks HM behaviour. There was no objection to that. Post-meeting note: Further testing to confirm the described results was conducted after the meeting by the software coordinator (K. Sühring). In the context of HM 16.2, the modified referencing structure reportedly showed an average 0.3% loss relative to the ordinary random access configuration for the HM, with peak loss (on the sequence "SteamLocomotive") being 2.1% (Main) or 2.2% (Main 10). There was a reported general tendency for higher resolutions to have somewhat larger losses than for lower resolutions. Since the reported loss, while still minor, was larger than what was expected from the discussion at the meeting, the example random access configuration for the HM was not replaced – pending future discussion to determine whether the measured loss is acceptable. (Only the example temporal scalability configuration was replaced for the HM distribution.) 13.0.0.1.1.1.1.1.338JCTVC-S0243 Cross-check of S0090 (On referencing structure supporting temporal scalability) [W. Wang, M. Xu, Z. Ma, H. Yu (Huawei USA R&D)] [late]
(Consideration of this topic was chaired by GJS on Thursday 10-23 p.m.) This information document provides notice of recent updates to conversion tools that feature scripts written in ctlrender (Colour Transform Language) by the Academy of Motion Pictures Arts and Sciences (AMPAS) that performs video conversion tasks, in particular High Dynamic Range (HDR) & Wide Colour Gamut (WCG) signals. New scripts added since the Sapporo meeting in July 2014 utilize ctlrender’s internal tone mapping algorithm to render Standard Dynamic Range /Standard Colour Gamut (SDR/SCG) signals, conforming to BT.709, from larger colour volumes in higher bitdepth HDR/WCG signal containers such as PQ (SMPTE ST 2084) / BT.2020. This was an information document to inform participants of the availability of these software tools. It was presented only briefly, but is available for study. 13.0.0.1.1.1.1.1.340JCTVC-S0206 On lambda-domain Rate Control [J. Wen, M. Fang, M. Tang (Tsinghua Univ.)] [late] (Consideration of this topic was chaired by GJS on Thursday 10-23 p.m.) This contribution reports several problems in the R-lambda model based rate control which was proposed in JCTVC-K0103. It is noted that R-lambda model based rate control fails to achieve the expected performance in some cases and the model may provide an inaccurate lambda without the clipping is applied, although it performs well on the test sequences. The presentation was requested to be uploaded. The contribution asserts that the model does not seem to be working well for some test clips. It suggests that further study be performed to improve upon the current behaviour – which actually reported a negative correlation between the lambda predicted by the model and that which should have been applied. Further experiments would be needed to confirm/determine what needs to be done. Further study was encouraged.
|