Joint Collaborative Team on d video Coding Extension Development of itu-t sg 16 wp and iso/iec jtc 1/sc 29/wg 11
Yüklə 1,87 Mb.
|
- Bu səhifədəki naviqasiya:
- 8Alternative depth formats (1)
- 9Non-normative contributions (4)
7.4Other (3)(chaired by A. Vetro) JCT3V-G0071 The removal of quad-tree limitation for depth maps [S. Yoo, J. Heo, S. Yea (LGE)] This proposal has already been discussed in the context of CE2. No action, but proponents may study further. JCT3V-G0202 Cross check of removal of quad-tree limitation for depth maps (JCT3V-G0071) [J. Y. Lee, C. Kim (Samsung)] [late] JCT3V-G0151 AHG13: Results with quarter resolution depth map coding [S. Shimizu, S. Sugimoto (NTT)] This contribution reports the result of quarter resolution depth map coding on MV-HEVC. In the simulation, VSO was activated for quarter resolution depth map coding with assuming the nearest neighbour upsampling would be performed during view synthesis. Experiments reportedly show that about 3.38% bitrate reduction and 6.22% bitrate increase for coded and synthesized views, respectively, with saving about 30% decoder runtime. The main benefit of reduced resolution depth is to reduce the decoder run time, or other implementation factors such as memory or throughput. It was observed that VSO gives significant improvement and uses a simple nearest neighbor upsampling. It would be desirable to use a more complex up-sampling filter to interpolate the depth values used for rendering, but this would need to be consistent with that used for VSO. It might be beneficial if VSO used a more complex up-sampling process. The contribution also analyzes the impact of reduced resolution depth on the coding tools in 3D-HEVC. There would be substantial work involved to support reduced resolution depth with all the 3D-HEVC coding tools. The contribution recommends supporting reduced resolution coding of depth in the software. The MV-HEVC specification supports reduced resolution depth since auxiliary pictures can have a different spatial resolution than the primary pictures. There were some reservations to adopt this into the software since it doesn’t provide a benefit in terms of coding efficiency. Also, there was some concern that there might be a desire to add other scaling ratios and make the software more complex. There was some interest to integrate into the software so that this capability could be studied further. It was suggested that the software might be integrated into a branch of the software, or simply shared with interested parties. It was noted that even the branch would need to be maintained and there were no volunteers to do this. As a result, it was agreed to simply share the software with interested parties. It was also agreed that the mandates of the AHG13 have been completed, and there is no need to continue the activity.
This document provides technically improved texture and depth view packing SEI message in 3D-AVC. The document is improved based on the editor’s comments at the 6th JCT-3V meeting. The new version of the SEI message has new syntax elements for homography mapping data, expressed via camera parameters. The syntax of the camera parameters is similar to the existing camera parameter SEI message (different in the length of the exponent of the float representation). Discussion whether the homography mapping could be expressed more compactly. However, the camera parameters are further needed to determine the pixel shifts from the depth values in the synthesis process. It is suggested that the camera parameters should be identical in syntax and semantics to the existing SEI message Position of the residual views in the frame packing should be changed to: R1 at 0,0 R2 at W/2, 0 R3 at 0, H/2 R4 at W/2, H/2 If there is only one residual view, it will be at R1, if there are two, they are at R1 and R2 etc. It does not matter where the physical views are, and how they are associated with the residual views, as the camera parameters are describing that. Remove Figures J1 and J2, and do not associate the base view with #3. The changed document was uploaded. The parent bodies gave advice to defer further consideration for the next meeting, also identifying relation with MPEG’s FTV exploration. 9Non-normative contributions (4)9.1Encoder optimization (4)(chaired by A. Vetro) JCT3V-G0079 Improvement on AMP fast mode decision [X. Chen, X. Zheng, Y. Lin, J. Zheng, (Hisilicon)] In this document, we constrained AMP_MRG using condition and test results of asymmetric motion partition (AMP) are provided. Three methods average gain are 0.36%, 0.34% and 0.3% with coding time 3.94%, 2.57% and 1.81% increase, respectively. The method was also proposed for HEVC single layer coding, but has additional benefit for multiview video coding. Currently, there is a fast estimation process to determine the AMP merge mode which incurs coding loss. The proposed method maintains the fast estimation, but does not lose as much. It was considered desirable by one expert to use the same base layer encoding techniques as HM; it was suggested that the method should be proposed to JCT-VC. it was reported later that for the case of HM the encoding runtime increases by 10%, giving a similar gain. For 3D HEVC, it is increasing the encoding runtime by 2%. No action.
Bugfix of VSD (View Synthesis Distortion) calculation is proposed to correct the bitDepth value assignment for VSD calculation. After bugfix, 0.1% BD-bitrate saving is achieved on synthesized view under CTC. Additionally a modification of VSD calculation is proposed and 0.1% BD-bitrate saving on view 1 and view 2, and 0.3% BD-bitrate saving on synthesized view are achieved under CTC. It was remarked that the bug fix seems like an obvious thing to do, and the proposed weighting function appears to provide a good gain. There was a question on whether the function was optimized to the test set – it was not. The same optimization can be applied for both MV-HEVC with depth as well as 3D-HEVC. Offline discussion has taken place to understand technique and confirm cross-check. Experts understand the proposed fix and improvement and support their inclusion in the software.
Yüklə 1,87 Mb. Dostları ilə paylaş:
|