Joint Collaborative Team on Video Coding (jct-vc) Contribution

6.5Interlaced scan and field-based video coding

JCTVC-M0441 HM software modifications for interlaced coding [Zineb Agyo, Jerome Vieron, Jean-Marc Thiesse] [late]

There was action taken at the last meeting to improve HM handling of interlaced video content (ref. JCTVC-L0187). Follow-up work has been ongoing together with the software coordinator. Further refinement is delegated to the software coordinator.

6.6Non-normative: Encoder optimization, decoder speed improvement, post filtering, loss concealment, rate control

6.6.1Rate control

JCTVC-M0036 Adaptive bit allocation for R-lambda model rate control in HM [B. Li, H. Li, L. Li (USTC)]

Improvements of the R-lambda rate control in HM. Reports that fixed-QP encoding benefit relative to rate control is in the range of 1-2%.

Also includes reported bug fix for intra rate control.

The initial version of the contribution did not include intra. A modification was provided later that included intra, in response to M0257.

(Initially reviewed in BoG M0460 chaired by A. Duenas.)

This document presents an adaptive bit allocation algorithm to improve the R-lambda model rate control algorithm in HM. With the proposed adaptive bit allocation algorithm, when targeting the HM-10.0 anchor bitrate, the coding efficiency is improved by about 2.9%, 2.9%, 1.9%, and 2.0% for RA-Main, RA-Main10, LB-Main, and LB-Main10 respectively (Class F sequences excluded). The average performance gap between enabling rate control algorithm and HM-10.0 anchor is about 2.4%, 2.3%, 1.9%, and 1.9% for RA-Main, RA-Main10, LB-Main, and LB-Main10 respectively (Class F sequences excluded). The bitrate accuracy is also improved a little.

This document also presents several modifications to support intra picture rate control better. This document applies a unified model and parameter updating method for both intra and inter pictures. With the modifications for intra coded pictures, the average Y BD-Rate gap between the R-lambda model rate control algorithm and HM-10.0 anchor are 0.8%, 0.7%, 2.5%, 2.4%, 1.8%, 1.7%, 1.1%, and 0.9% for AI-Main, AI-Main10, RA-Main, RA-Main10, LB-Main, LB-Main10, LP-Main, and LP-Main10 respectively. The average sequence level bit rate error are 0.00%, 0.00%, 0.18%, 0.19%, 0.07%, 0.11%, 0.09%, and 0.10% for AI-Main, AI-Main10, RA-Main, RA-Main10, LB-Main, LB-Main10, LP-Main, and LP-Main10 respectively.

The BoG recommended adoption on future versions of the reference software model for HEVC of the enhancements that are not related to intra picture that are presented on this contribution and thanked USTC for their work on this contribution and the improvement of the reference software model for HEVC.

Decision (SW): BoG recommendation confirmed (not high priority).

JCTVC-M0434 Cross-check of JCTVC-M0036 [J. Xu (Microsoft)] [late]
JCTVC-M0037 Rate control by R-lambda model for SHVC [L. Li, B. Li, H. Li (USTC)]

Reports on some issues and needs for harmonization in SHM with techniques in HM.

(Initially reviewed in BoG M0460 chaired by A. Duenas.)

This contribution extends the R-λ model based rate control algorithm in HEVC to SHVC. The proposed algorithm was implemented in SHM-1.0 for both the ref_idx framework and Intra_BL framework. When targeting the bitrate of enhancement layer of SHM-1.0 anchor, the average bitrate errors (the difference of the target bitrate and the actual bitrate) of the proposed algorithm are about 0.01%~0.14% for different cases, the average luma BD-rate compared to the anchor is about 2.5%~4.4%.

It was noted that they are potential improvement to be done to the current R-λ model based rate control algorithm in SHVC and this proposal address some of them.

The BoG recommended adoption of this contribution on future versions of the reference software model for SHVC and thanked USTC for their work on this contribution the improvement of the reference software model for SHVC.

Decision (SW): BoG recommendation confirmed.

JCTVC-M0415 Crosscheck for Rate control by R-lambda model for SHVC (JCTVC-M0037) [W. Zhang, Y. Chiu (Intel)] [late]
JCTVC-M0257 Intra Frame Rate Control Based on SATD [M. Karczewicz, X. Wang (Qualcomm)]

Proposes rate control enhancements for intra coding.

(Initially reviewed in BoG M0460 chaired by A. Duenas.)

In this contribution, an R-λ model based rate control scheme for intra frame/slice is proposed for HEVC. The proposed rate control algorithm is implemented on HM-10.0. Compared with the existing rate control algorithm in HM-10.0, it is asserted that the proposed algorithm provides more accurate matching of target rate while maintaining similar coding efficiency.

Presentation not yet uploaded.

It was mentioned that several of the proposed ideas have been also presented on JCTVC-M0036.

The differences between this proposal and JCTVC-M0036 were discussed, and it was mentioned that they do actually have many similarities and that JCTVC-M0257 has some enhancements in addition to what is proposed on JCTVC-M0036.

The BoG recommended adoption of this contribution on future versions of the reference software model for HEVC and thanked Qualcomm for their work on this contribution.

In JCT-VC review, it was commented that there may concern if the scheme shifts bits away from the enhancement layer. The cross-checker indicated that this was not the case – that more bits are actually allocated to the enhancement layer by the scheme.

Decision (SW): BoG recommendation confirmed (for integration after M0036).

6.6.2Encoder optimization

JCTVC-M0259 Encoder Optimization for SHVC: Enhancement Layer Lambda Refinement in RA Configurations [X. Li, J. Chen, M. Karczewicz (Qualcomm)]

(Initially reviewed in BoG M0460 chaired by A. Duenas.)

For SHVC, R-D optimization enhancement for EL. 0.8% gain was reported for TextureBL and RefIdx for RA.

In this proposal, an enhancement layer encoder optimization method is proposed for SHVC RA configurations. Lagrange multiplier in rate-distortion module of enhancement layer is refined based on temporal level, resolution ratio and QP gap between base and enhancement layer. It is reported that on average 0.8%, 1.1% and 1.2% BD-rate reduction was obtained for Y, U, and V components in both IBL and RefIdx frameworks.

No presentation uploaded.

The BoG recommendeds adoption of this contribution on future versions of the reference software model for SHVC and thankeds Qualcomm for their work on this contribution and thanks Qualcomm for their work on this contribution the improvement of the reference software model for SHVC.

Decision (SW): BoG recommendation confirmed – enable by default.

JCTVC-M0343 Cross-check for Encoder Optimization for SHVC: Enhancement Layer Lambda Refinement in RA Configurations [E. Alshina] [late]

6.6.3Software development

JCTVC-M0398 Multiview and multiview-video-plus-depth support in SHM [A. Hallapuro, Y. Yan, D. Rusanovskyy, M. M. Hannuksela] [late]

Presented in closing plenary.

In the JCTVC-L/JCT3V-C meetings, the specifications of MV-HEVC and SHVC were aligned to use a common basis. This contribution presented a software model (SM) that implements the MV-HEVC specification on top of the SHM v1.0 software and hence is asserted to provide functionality for collaborative development on aligned specifications for spatial and multi-view extensions of HEVC. In addition to quality/spatial scalability and multiview coding, the proposed SM can be used for Multiview Video plus Depth (MVD) data, mixed resolution MVD data (where the depth views have a different spatial resolution than the texture views) and asymmetric multi-view video data (where a base view has a lower spatial resolution than non-base views). All these configurations of SM are reportedly still aligned with the MV-HEVC and SHVC specifications.

It is proposed to adopt the presented SM as a Test Model for collaborative development on aligned specifications for spatial, multi-view and MVD extensions of HEVC.

It was suggested to consider using this software as the primary SHVC software basis moving forward. However, the software had not been made available yet. The contributor volunteered to upload the software as a revision of the contribution.

Decision (SW): The SHM software development AHG was tasked with determining whether this is feasible and, if so, proceeding on that basis.

It is understood that normative technical proposal adoptions entail contribution of related software and work to integrate the software properly. It was asked whether this extends to multiple codebases, and concluded that for now participants may not be fully obligated to contribution to codebases other than the primary codebase for the project related to the proposal.