International organisation for standardisation organisation internationale de normalisation

Alternative depth formats

At the last JCT-3V meeting, the Alternative Depth Info SEI message was adopted for inclusion in the specification of the 3D-HEVC Test Model. To address corresponding editorial comments available in the specification of the 3D-HEVC Test Model 4, this document specifies the persistence scope of Alternative Depth Info SEI message and provides an update of the syntax and the semantics description text without changing the functionality of the SEI message.

A new flag is introduced that can deactivate the SEI message again.

For persistence, add specification that once acivated, the SEI message is valid up to the end of the sequence (i.e. as long as the current SPS is valid), unless deactived.

(xyz_plus1 should be changed to + 1.)

W and H should be exclusively even numbers.

Decision: Adopt

(Changes to be checked with editor.)

15.0.1.1.1.1.1.1.293JCT3V-E0105 AHG8: Revision of Depth View Information SEI Message in 3D-AVC DT6 [T. Senoh, K. Wakunami, Y. Ichihashi, H. Sasaki, K. Yamamoto (NICT)]

Responding to the editor’s comments on the semantics description of depth view information SEI message in 3D-AVC DT6, a revised text is proposed.

Editorial improvements and clarifications are suggested in the contribution on items as requested by the text editor (MH). No technical of the SEI message itself.

Decision (Ed.): Integrate into the next draft of 3D-AVC in coordination with the editor.

15.0.1.1.1.1.1.1.294JCT3V-E0106 AHG8: Revision of Global View and Depth Format Description for Depth View SEI Message [Takanori Senoh, Koki Wakunami, Yasuyuki Ichihashi, H. Sasaki, K. Yamamoto (NICT)]

According to the following mandate of AHG8, this contribution revises Global View and Depth (GVD) format description for Depth View SEI message.

• 
  Improve the description of global view and depth format description in 3D-AVC Test Model.
  

Note: Current ATM does not have a description of view synthesis procedure (whereas the HTM description has such a description). It is also agreed that the ATM editor shall include a description of view synthesis in the ATM text (can just be copied from HTM).

8. Non-normative contributions

A rate control algorithm is an essential part of the practical video encoder for services such as VoD (Video on Demand), video conference, etc. The latest reference software of HEVC, HM-11.0, includes two rate control algorithm s based on URQ (Unified Rate Quantization) and R-lambda model. In contrast, the current test model for multi-view video coding, HTM7.0_r1, based on HEVC does not include a rate control algorithm. As the base view of HEVC-based multi-view video coding is compatible to HEVC, the rate control algorithms in HEVC can be used for both the base view and extended views immediately. However, a rate control algorithm for the extended views can be improved rather than the current rate control algorithms by using the inter-view correlation. Before thorough analysis about a rate control algorithm for the extended views, the former rate control algorithms in HEVC (HTM-DEV0.3) should be implemented in HTM7.0_r1 for the performance evaluation. In this proposal, based on the current rate control algorithms in HTM-DEV0.3 on HTM7.0_r1 to extend those single-view rate control algorithm to multi-view video coding and propose the view-wise target bitrate allocation and inter-view MAD prediction. Those proposed algorithms are compared the performance of the rate control algorithms for each view to the anchor (HTM7.0_r1 with rate control under CTC) under the CBR (Constant Bit Rate) condition. From the performance evaluation results, there were 0.24% bitrate accuracy degradation for all views and 2.67%, 2.91% BDrate gain in average for extended views by using R-lambda model-based rate control with the proposed algorithms. And 0.77% bitrate degradation for all views and 0.76% and 0.07% BDrate loss in average for extended views by using URQ model-based rate control with the proposed algorithms. In both cases of R-lambda and URQ model-based rate control with the proposed methods, the virtual buffer occupancy is more stable than the cases using URQ or R-lambda model-based rate control.

HM11 (and therefore also HTM-DEV2.0 which will be the basis of HTM8) has two different rate control methods: R-Lambda and URQ.

A new rate control is proposed here only for the dependent views (somehow using MAD from base view to predict the best QP of dependent view), which

• 
  when combined with R-Lambda gives 2.67/2.91% BR reduction for dependent views
  
• 
  when combined with URQ gives 0.76/0.07% BR increase for dependent views
  

15.0.1.1.1.1.1.1.296JCT3V-E0288 Cross-check on JCT3V-Inter-view MAD prediction for rate control of 3D multi-view video coding (JCT3V-E0227) [Y.J Lee, M.S. Lee, Y.S. Heo, G.H. Park (KWU)] [late]

9. Complexity assessment

This contribution provides a case study of complexity assessment on simplified depth coding (SDC) in 3D-HEVC and the assessed complexity included number of operations, data storage requirement, and data transfer rate. The results in this proposal can be considered as anchor complexity to assist JCT3V to evaluate the proposals coming from several proponents in a fair manner. This proposal also presents the experimental result without SDC coding tools. The results show that the coding performance will be degraded without SDC in both CTC and AI configuration (0.29% and 0.28% for coded and synthesis BD-rate at CTC configuration and the coding loss is 0.96% and 0.84% for coded and synthesis BD-rate at AI configuration) and the decoding time will be increased at AI configuration without SDC (18.9% decoding time increasing).

Assessment of memory complexity may require more investigation.

Continue study in AHG.

10. 
    
    Directory: sc29 -> open -> 29view
    29view -> International organisation for standardisation organisation internationale de normalisation
    29view -> International organisation for standardisation organisation internationale de normalisation
    
    
    
    Download 8.47 Mb.
    
    Share with your friends: