5.2.1.1.1JVT-Z030 ( Prop 2.2/3.1) [Y. S. Ho, K. J. Oh, C. Lee (GIST)] Regional disparity derivation for MVC motion skip mode
This document described a method of regional disparity derivation for motion skip mode. The current motion skip mode in JMVM utilizes a global disparity vector of 16-sample precision to find the position of the corresponding macroblock for the current macroblock. However, since the multi-view scene consists of several objects and each object has its own disparity value, it was asserted that the global disparity is not enough to cover the disparity of the whole image. It was proposed to use regional disparities, instead of the global disparity, for the motion skip mode. The proposed scheme generates the disparity map for each anchor frame considering its motion vectors and then derives disparity maps for non-anchor frames using both forward and backward disparity maps. The temporal movement is also considered. Compared to JMVM 6.0, the proposed scheme reportedly achieved a similar coding gain with the previous scheme.
Results were reported for 5 sequences, 2 GOPs. Relative to the current motion skip mode, approximately the same coding performance was reported.
The proponent suggested combining with residual prediction and suggested that better results might be obtained that way.
Contribution noted.
5.2.1.1.2JVT-Z031 ( Prop 2.2) [J. H. Park, B.H. Choi (KETI)] MVC motion skip mode with residual pred
This document proposes a prediction structure for MVC which is reportedly a combination of motion skip mode (per JVT-Y058) and residual prediction. In terms of residual prediction, it was claimed that the proposed method is very similar to the residual prediction technique of SVC. The proposed method uses an integer precision global disparity vector and derives a disparity vector of motion skip mode from the global disparity vector by a shift operation. For such use, it was asserted that a smoothing filter would be needed to reduce boundary artifacts, but investigation of such a filter reportedly could not be finalized due to lack of time. The proposed method reportedly showed some gain without a smoothing filter for "dense" sequences. The number of test sequences was limited – 1.4% average gain on set of 5 sequences. The contribution recommended the creation of a CE on this topic.
Remark: Adds substantial complexity (e.g. searching in encoder for integer-precision GDV value, residual handling and storage in decoder).
Contribution noted.
5.2.1.1.3JVT-Z032 ( Prop 2.2) [J. H. Park, B.H. Choi (KETI)] Clarification of motion_skip_enable_flag
This document requested a change of the conditions relating to the slice header syntax element motion_skip_enable_flag, which was newly introduced from JVT-Y207. This contribution suggested to change the conditions on the presence of the motion_skip_enable_flag so that it is not sent when it is not used, and to otherwise structure the syntax in a more logical fashion in relation to that syntax element.
JVT Decision: Adopted (conditioned on whether we actually will use the feature that this is refining the design of).
