International organisation for standardisation organisation internationale de normalisation
Yüklə 7,38 Mb.
|
- Bu səhifədəki naviqasiya:
- Alternative depth formats (2)
AVC related (2)14.1.1.1.1.1.1.1.251JCT3V-F0152 Depth-based motion vector prediction in 3D-AVC [J. Y. Lee, M. W. Park, H.-C. Wey, B. Choi, Y. Cho, C. Kim (Samsung)] The current 3D-AVC supports both texture- and depth-first coding orders. In the depth-first coding, 3D-AVC achieves the significantly high coding performance compared to MVC+D. However, it was reported that the performance of 3D-AVC is very small in the texture-first coding. One of the reasons why the previous 3D-AVC has the low coding efficiency in the texture-first coding is that the depth-based motion vector prediction (DMVP) is not supported. Since the coded depth map for the motion vector prediction of the associated texture image is not available, DMVP is turned off. In order to improve the coding performance in the texture-first coding, MB-level NBDV (E0136) was adopted at the last meeting. Hence, the current 3D-AVC includes the two different motion vector prediction methods – DMVP for the depth-first coding and NBDV for the texture-first coding. As a follow-up proposal of E0148, this contribution proposes DMVP in the texture-first coding, instead of NBDV, to unify the different motion vector prediction method in the texture- and depth-first coding. As the first option, the proposed method employs the disparity converted from the global depth value (GDV, E0035) and the updated disparity for the first block and the remaining ones, respectively. As the benefit, motion vectors in the dependent views are not needed to be stored unlike NBDV. As the second option, if a disparity vector in the collocated block in the previous frame is available, the proposed method uses it in the same manner as NBDV. If not available, GDV and the updated disparity are employed. Since both the options employ DMVP, the proposed method can unify the motion vector prediction part in 3D-AVC. Implementation of the proposal would lose efficiency (approx. 1% bit rate increase), if DMVP is operated “as is” instead of NBDV. If DMVP would be changed to use the MV of the previous frame in texture first coding, the performance would be almost identical, however this would imply a modification of a tool at a very late stage of the standardization process and the intended goal of unification would not be achieved. The claimed advantage of unification does not justify such a loss, and a technical change without benefit would not be desirable. No action. 14.1.1.1.1.1.1.1.252JCT3V-F0179 Crosscheck on Depth-based motion vector prediction in 3D-AVC (JCT3V-F0152) [S. Shimizu, S. Sugimoto (NTT)] [late]
Note: bug fix for 3D-AVC, misleading “CE3” title The disparity-based skip/direct mode directly inherits the motion information, i.e., reference index(es) and motion vector(s), of a corresponding block in the base view as the motion information of the current block. However, an inherited reference index may exceed the maximum reference index of the current block, which may cause the encoder and decoder to crash. Therefore, it is proposed to add one additional check to see if the inherited reference index is valid. If the reference index is valid, the reference index and MVs of the corresponding block are directly inherited. Otherwise, the inherited inter-view motion information is viewed as unavailable. This bug-fix does not occur under common test conditions.
Reference index(es) and motion vector(s) of the corresponding block in the base view (or reference view) are directly inherited by the current block. However, the inherited reference index may exceed the maximum reference index of the current block and the encoder or decoder may crash due to this invalid reference index. The suggested solution is checking whether the inherited reference index exceeds the maximum reference index of the current block. When the inherited reference index exceeds the maximum reference index, the inter-view motion predictor is viewed as unavailable. Decision: Adopt.
14.1.1.1.1.1.1.1.254JCT3V-F0087 2D Backwards Compatible Centralized Color-Depth Packing [Jar-Ferr Yang, Hung-Ming Wang, Yi-An Chiang (NCKU)] [late] The most popular frame compatible formats for transmitting stereoscopic views are side by side (SbS) and top and bottom (TaB) formats. However, there are still no related standard packing formats for videos with color and depth information. The proposed method targets on a 2D Backwards Compatible Centralized Color-Depth Packing (CCDP) for videos with color and depth information. The proposed format is compatible with ordinary 2DTV, which means it can be directly displayed without extra pre-processing. When we want to display the 3D contents, the color and depth information can be extracted easily with comparable quality and generate the desired 3D contents through rendering technique. PSNR values against simulcast shows no performance drop. 14.1.1.1.1.1.1.1.255JCT3V-F0088 2D Backwards Compatible Centralized 2 view Packing [Jar-Ferr Yang, Ke-Ying Liao, Chen-Yao Yeh, Ya-Han Hu (NCKU)] [late] This contribution develops a new packing format, called the centralized 2-view (C2V) 3D video packing format that provides a more unrestricted and comfortable vision experience for traditional 2D display. With C2V format, the users can perceive the more comfortably TV programs without any un-packing process, while the traditional 2D TV receiving 3D video contents. Similar to the first contribution, but additionally uses checkerboard subsampling of left and right videos. Performance drop compared to simulcast. Main intention is to achieve viewing comfort on 2D display without cropping. No concrete proposal for a standardization action. (Note: So far, frame packing formats have been defined per SEI message, which are outside the scope of normative decoder behaviour) No action to be performed at this point.
|