Organisation internationale de normalisation

SCC Other (6)

13.0.0.1.1.1.1.1.321JCTVC-S0075 Copy Mode for Static Screen Content [T. Laude (Leibniz Universität Hannover)]

(Consideration of this topic was chaired by G. Sullivan on Saturday 10-18 p.m.)

This contribution presents a copy mode which reportedly aims at the coding of static screen content. In particular, it is reported that the sample values of a block coded with the copy mode are reconstructed by copying the sample values from the corresponding block at the same position in the closest reference picture. Furthermore, it is asserted that the copy mode is only applied on CTU level. The contribution states that Y/G BD-rate changes of {−0.2%, 0.0%, −0.5%, −0.6%, 0.0%, 0.0%, −0.2%, −0.3%, −0.6%, −1.0%, 0.1%, 0.0%} and {0.0%, 0.0%, −0.2%, −0.1%, 0.0%, 0.0%, 0.0%, −0.1%, −0.1%, −0.1%, 0.1%, 0.0%} are achieved for LD and RA compared to SCM-2.0 under SCC common test conditions.

The contribution proposes a context-coded flag (with one context) prior to the skip flag. When the flag is equal to 1, the CTU is copied from the picture with the closest POC. (This conceptually differs slightly from the usual collocated picture which we normally signal explicitly.)

The reported gain ranges from −0.1 to 1.0% per sequence category.

It was commented that there may be an R-D optimization effect in the measured results.

It was suggested to try setting the max number of merge candidates to 1.

It was asked whether the percentage gains are in cases where the compression ratio is already very high.

It was asked whether there may be PSNR clipping effects.

Further study was encouraged to check the result and determine whether there is more than an RDO effect or a measurement phenomenon involved in the measured results.

13.0.0.1.1.1.1.1.322JCTVC-S0085 Adaptive motion vector resolution for screen content [B. Li, J. Xu, G. Sullivan, Y. Zhou, B. Lin (Microsoft)]

(Consideration of this topic was chaired by JRO on Thursday 10-23 a.m.)

This document proposes and reports performance measurements for disabling fractional pixel motion compensation in high-level syntax, as previously proposed in JCTVC-P0277, JCTVC-Q0155 and JCTVC-R0106. For many screen contents, the spatial displacements are of full pixel and the encoder can be notified accordingly by the application system. Not to consider fractional-pel displacement not only save bits on motion vectors, it also reduces the complexity of both the encoder and decoder. The proposed modification does not change the syntax or parsing process. A method to determine the MV resolution in SCM-2.0 encoder is also developed. The modification is shown to improve coding efficiency as much as 7.9% for some screen content. For RGB text and graphics with motion (TGM) 1080p sequences, 2.7% and 4.0% bit saving on average is obtained for RA and LB lossy coding. The experimental results show that most of the coding gain could be obtained without increasing the encoding running time.

The contribution proposed to include a slice-level flag.

The encoder decision for some of the presented results is multi-pass. It was shown that for the TGM class an encoder-only solution does not give the same benefit (0.4% vs. 3.0% for RA, 0.5% vs. 4.1% for LDB). With a single-pass coding method, the gain is still 2.7% and 4.0% for RA and LD (lossy).

The decoding of the MV is unchanged. The suggested change is to re-interpret the MV (<<2) before the MC stage.

TMVP is not changed.

It is pointed out that in real applications a fast algorithm may not be as simple to build as in the reference software. The proponents suggest that there could also be situations where the encoder has knowledge about the presence of only integer motion.

The additional decoder complexity would be an additional condition check and two shift operations at the PU level. This additional complexity appears to be minor, and a good tradeoff with the achieved compression gain. In the context of SCC, new designs are likely to be necessary anyway at the PU level (e.g. when using same circuits for MV and BV dependent compensation).

Two options were proposed: Only switching quarter/full, or switching quarter/half/full pel precision.

According to several experts' (and the proponents') opinion, the first option (quarter-pel/full-pel switching only) is the preferred solution; no advantage is evident to also enable the half-pel option.

Decision: Adopt (first option – integer/quarter pel adaptation with enabling flag at slice level, and feature enabling flag at SPS).

The method (with fast encoder decision) shall be used in CTC. It was recommended to perform further study whether this is meeting the requirement of latency in LD case.

13.0.0.1.1.1.1.1.323JCTVC-S0239 Cross-check report of JCTVC-S0085 on Adaptive motion vector resolution for screen content [K. Rapaka (Qualcomm)] [late]

13.0.0.1.1.1.1.1.324JCTVC-S0187 Non-SCCE: SCC with extended LCU size [D. Jiang, X. Zhang, Y. Wu, Z. Wang (Lenovo)]

(Consideration of this topic was chaired by GJS on Wednesday 10-22 a.m.)

This contribution proposes an extension of current coding framework by extending the largest coding unit size. Since there are more repeated and flat blocks in SCC testing sequences, conspicuous BD-rate gain could reportedly be achieved by using an enlarged largest coding unit.

Experimental results reportedly show the proposed approach provides bitrate effects of −2.8 %, −3.6 % for AI 1080p text and graphics RGB and YUV categories respectively relative to the SCM2.0 anchor. At the same time, the decoder complexity is reportedly reduced for AI, LB and RA. The encoding time is increased with 142%, 147% and 139% for AI, LB, and RA lossy configurations, respectively.

Proposes 128x128 max CTU size and PU size. Max TU still 32x32.

Gain for AI is from 0.1 to 3.6% depending on category – a bit less in RA and LD.

Presentation table columns had a labelling mixup.

40-50% coding time increase.

It was remarked that the implications on implementation would be substantial – both in terms of the amount of memory that must be accessed together at high speed and in terms of the ability to adapt a design that was previously used for v1 implementation.

It was remarked that there is an interaction with other issues such as non-normative encoder decisions.

It was remarked that the method probably has the most benefit on material that is otherwise relatively easy to encode to a high compression ratio.

The information was appreciated, and it may be the case that a larger CTU size could improve coding efficiency; however, the implementation burden of this may be too much.

Further study was encouraged to try to identify the sources of the gain and determine whether some adjustments to the design (e.g., to palette mode coding or IBC or non-normative decisions or merge operations) might be able to yield similar results.

13.0.0.1.1.1.1.1.325JCTVC-S0283 Crosscheck of JCTVC-S0187 on SCC with extended LCU size [X. Li, Y. Chen (Qualcomm)] [late]

13.0.0.1.1.1.1.1.326JCTVC-S0272 Intra Reference Prediction by Cross-Component Prediction [K. Kawamura, S. Naito (KDDI)] [late]

(Consideration of this topic was chaired by G. Sullivan on Saturday 10-18 p.m.)

Since HEVC version 1 standardization, linear-model chroma prediction (LM chroma) was studied. However, a cross-component dependency by the LM chroma was suggested to be unacceptable for a hardware implementation. In this contribution, chroma intra reference samples are predicted from luma samples by using cross component prediction. The dependency in this contribution is reportedly the same as that by the cross-component prediction in RExt. BD-rate impacts for RGB/YUV videos under lossy coding with G/Y, B/U, R/V of the all intra condition were reported.

The contribution proposes a new intra prediction mode in which an intra prediction block for chroma is predicted using a linear model applied to luma boundary sample values, where the parameters of the linear model are derived at the decoder side.

The reported test results showed some loss in luma and gain in chroma – e.g., 0.0-0.6% loss in luma and 0.0-2.9% gain in chroma, depending on the sequence category. The biggest gains are in YUV cases, where not so much bit rate may be expended on chroma. Measurements would need to be refined to avoid the question of whether this is just bit rate allocation shifting from luma to chroma.

No action taken on this.

2.