Sub-Block Motion Candidate Derivation Using Spatial Neighbors
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current video coding methods face challenges in efficiently processing sub-block motion candidates, leading to increased complexity and reduced coding performance, particularly in handling affine and temporal motion vector predictions across different video standards.
Innovation Solution
The proposed method simplifies sub-block motion candidate lists by using adaptive ordering and selective inclusion of motion candidates, such as ATMVP and affine candidates, within a sub-block based merge list, excluding block-level prediction candidates, and utilizing a history-based motion vector prediction table for improved prediction accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If traditional video coding methods are used for sub-block motion candidate processing, then comprehensive motion prediction coverage is achieved, but hardware implementation complexity increases
Solution Approach 1:
The motion candidate list is segmented into different types (spatial, temporal, HMVP) and organized in a hierarchical structure. The method divides the candidate selection process into multiple stages: first selecting from spatial neighbors, then temporal neighbors, and finally HMVP table, with each stage having specific selection criteria. This segmentation reduces the computational complexity of evaluating all candidates simultaneously while maintaining comprehensive coverage through the multi-stage approach.
Solution Approach 2:
Different selection criteria and priorities are applied to different types of motion candidates based on their local characteristics. Spatial neighbors are prioritized when available, with specific weighting given to left and above neighbors. Temporal candidates are selected based on reference picture list configuration. HMVP candidates are used as fallback options. This local quality approach optimizes the selection process for each candidate type's specific properties rather than using a uniform selection method.
2Productivity
If comprehensive motion candidate lists are maintained for all block types, then coding performance is improved, but processing time and complexity increase
Solution Approach 1:
The motion candidate lists are prepared and organized in advance during the encoding process. Spatial, temporal, and HMVP candidates are pre-identified and stored in their respective lists before the actual motion compensation operation. The HMVP table is maintained and updated during picture decoding, making candidates readily available. This preliminary action eliminates the need for real-time computation of all candidate lists during motion compensation, significantly reducing processing time while maintaining comprehensive candidate coverage.
Solution Approach 2:
The motion candidate selection process is made dynamic and adaptive based on the specific coding context. The method dynamically adjusts which candidate types are selected and in what order based on availability flags, picture type (I/P/B slices), and reference picture list configuration. For example, temporal candidates are selectively included based on collocated block availability, and HMVP candidates are dynamically added to the list. This dynamic approach optimizes processing time by avoiding unnecessary candidate evaluations while maintaining coding efficiency.
Data Source
AI summary
Devices, systems and methods for the simplification of sub-block motion candidate lists for video coding are described. In a representative aspect, a method for video processing includes determining, during a conversion between a current block and a bitstream representation of the current block, a temporal motion vector prediction candidate for a sub-block of the current block. The temporal motion vector prediction candidate is completely determined based on K neighboring blocks of the current block, K being a positive integer. The method also includes performing the conversion based on the temporal motion vector prediction candidate for the sub-block.