Joint Collaborative Team on Video Coding (jct-vc) of itu-t sg16 wp3 and iso/iec jtc1/SC29/WG11

5.4Non-normative: Encoder optimization, decoder speed improvement and cleanup, post filtering, loss concealment, rate control, other information (7)

JCTVC-V0034 Palette encoder improvements for the 4:2:0 chroma format [C. Gisquet, G. Laroche, P. Onno (Canon)]

(Consideration of this topic was chaired by GJS on Wednesday 10-21, 0915-–0925.)

During the 20th Meeting in Geneva, in February 2015, an extension of the palette-mode for the 4:2:0 chroma format has beenwas adopted. It operates by actually encoding 4:4:4 content, then discarding the superfluous chroma samples. It is asserted that the encoder does not properly account for this when producing the mapping of pixels to palette entries, causing a substantial loss in coding efficiency. The present contribution therefore suggests to weight the 4:4:4 chroma samples according to their position relative to the 2x2 grid of samples and the chroma format in several computations. It is reported that the non-normative changes provides a luma BDR gain of 3.2%, 2.4% and 1.3% for YUV TGM content in the AI, RA and LDB lossy configurations, respectively.

Decision (S/W): Adopt into SCM (enabled by default).
JCTVC-V0076 Cross-check of palette encoder improvements for the 4:2:0 chroma format (JCTVC-V0034) [V. Seregin (Qualcomm)] [late]
JCTVC-V0040 Performance of the SCM with the macro SCM_U0095_FAST_INTRA_ACT enabled [Y.-J. Chang, C.-L. Lin, J.-S. Tu, C.-C. Lin (ITRI)]

(Consideration of this topic was chaired by GJS on Wednesday 10-21, 0925-–0935.)

In the last meeting, JCTVC-U0095 proposed to disallow the TU splitting for the ACT-enabled intra prediction when the CU sizes are 64x64 and 32x32 in the encoder. It was reported that this method improves 4% saving of encoding time in AI-lossy coding conditions on top of SCM 4.0. It was also commented that it is desirable to test this method on top of adoption of JCTVC-U0106 (TU-based ACT). Therefore, the final decision was to adopt the method of JCTVC-U0095 into the next version of SCM and set the macro of this method as disabled by default for CTC. To understand the performance of this method on top of JCTVC-U0106, this contribution sets the macro of this method “"SCM_U0095_FAST_INTRA_ACT” " as on in SCM 5.2. The performance results reportedly show that, compared to the SCM 5.2 anchor, a 4% saving of encoding time on average in AI-lossy condition no matter how the condition of IBC search range is set. The reported amount of saving in SCM 5.2 is the same as was reported previously in SCM 4.0.

In RA, a 1% speed-up was reported.

The change was reported to involve only one or two lines of code.

Decision (S/W): Adopt into SCM (enabled by default). An algorithm description will also be included in the SCM algorithm document.

JCTVC-V0089 Cross-check of performance of the SCM with the macro “"SCM_U0095_FAST_INTRA_ACT”" enabled (JCTVC-V0040) [B. Li, J. Xu (Microsoft)] [late]
JCTVC-V0078 Improvement of coding efficiency for rate control under the constraint of HRD [Yong.-Jo. Ahn, Xiangjian . Wu, Donggyu . Sim (KWU)] [late]

(Consideration of this topic was chaired by GJS on Wednesday 10-21, 0935-–0945.)

This contribution proposes a modificaition to the coded picture buffer (CPB) and coding performance of the target bit saturation method that was proposed in JCTVC-U0132. To prevent the CPB from overflow and underflow, the target bit saturation method proposed in JCTVC-U0132 adjusts the target bit cost for a picture by using the fixed lower and upper bound of CPB fullness when it is expected to occur CPB overflow or underflow. In the performance evaluation of the contribution, some degradations of coding efficiency were observed for a few sequences. In this contribution, an improvement of the target bit saturation method is presented to achieve coding gain and to improve usages of CPB under the constraint of hypothetical reference decoder (HRD). To improve the coding efficiency of rate control under the constraint of HRD, this contribution proposes a target bit saturation for each picture using adaptive lower bound instead of a fixed lower bound. To determine the adaptive lower bound (L) for preventing CPB from underflow, the target bit cost for the next coded picture is estimated and considered. If the target bit for the next coded picture is expected to be high, the lower bound is set to be relatively high. Otherwise, the lower bound is set to be relatively low. It can reportedly guarantee to prevent CPB from underflow with adaptively using of CPB fullness.

A BDR loss of 0.8% was reported relative to the prior method, but the prior method was reported to exhibit excessive bit rate fluctuation effects.

It was commented that it is desirable to have a method that avoids excessive bit rate fluctuation, and that this is more important than minor coding efficiency effects.

The modification seemed like a minor adjustment of the prior scheme proposed by the contributor.

Decision (SW): Adopt (replacing the prior similar scheme for rate control).

JCTVC-V0088 Tile level rate control for multi-core platform [I. Marzuki, Y.-J. Ahn, X. Wu, W. Lim, D. Sim (Kwangwoon Univ.)] [late]

(Consideration of this topic was chaired by GJS on Wednesday 10-21, 0945-–0955.)

This contribution proposes a tile -level rate control for HEVC encoding on a multi-core platform on HEVC. Considering tiles as a tool for such a multi-core platform, practically it will typically degrade the video quality. This is reportedly due to the overlap threading process might usually occur in the multithreading tile level and it leads to the number of the encoder output bits being abnormally changed. Therefore, it is required proposed to consider tile level rate control for multi-core operation to control the parallel performance quality of tiles. However, the current R-Lambda model reportedly does not favour rate control on the tile level. Accordingly, this contribution proposes to inspect the tile level rate control for multi-core platform for HEVC.

Immediate action was not requested, and further study was encouraged. Certainly, improving the effective use of tiles in the reference software would be desirable, and the knowledge of such techniques would benefit the community.

JCTVC-V0096 Status report of rate control on 4k videos [J. Wen, M. Fang, Z. Wen (Tsinghua Univ.??)] [late]

(Consideration of this topic was chaired by GJS on Wednesday 10-21, 0955-–1005.)

This document reports the results of encoding with rate control on 4k videos content. It is was noted reported that rate control on 4k videos underperforms the constant-QP mode with a similar bit rate by 0.3 dB (more than 10% of the bit rate) on average and up to 0.6 dB.

This was an information document provided to encourage further study. The information was appreciated.

It was commented that a more recent version of the HM might do somewhat better in this study.

It was also noted that some gain was shown for the rate control usage for lower-resolution content.

It was suggested that using the usual Excel template in further study might provide more reliable results.

Further study was encouraged.

It was commented that testing rate control performance on SCC is also desirable.