Geometric Partitioning Merge Mode With Motion Vector Refinement
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The constraint on motion vector accuracy in the Geometric Partitioning Merge Mode (GEO) of the Versatile Video Coding (VVC) standard limits its coding efficiency due to restrictive support for motion vector combinations, preventing optimal partitioning flexibility.
Innovation Solution
The proposed solution extends the GEO mode by allowing additional motion vector refinement through the use of motion vector difference (MMVD) for coding units, utilizing new syntax elements to calculate motion vector differences based on distance and direction values for improved partitioning accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If Geometric Partitioning Merge Mode uses restrictive motion vector combinations, then implementation complexity is reduced, but coding efficiency deteriorates
Solution Approach 1:
The motion vector difference calculation is segmented into distance value and direction value components. This allows independent optimization of each component while maintaining overall simplicity. The distance value determines the magnitude of motion vector refinement, and the direction value determines the orientation, enabling flexible yet manageable complexity.
Solution Approach 2:
The patent introduces new syntax elements that change the parameters available for motion vector representation. By adding distance and direction values as separate parameters, the system expands the motion vector combination space from 30 fixed combinations to a continuous range of possibilities, significantly improving coding efficiency while maintaining a relatively simple implementation through parameter-based control.
2Productivity
If Geometric Partitioning Merge Mode supports more motion vector combinations, then coding efficiency is improved, but device complexity increases
Solution Approach 1:
The motion vector system transitions from static fixed combinations to dynamic calculated values. Instead of selecting from 30 predetermined motion vector combinations, the system dynamically calculates motion vectors using base motion vectors plus motion vector differences derived from distance and direction syntax elements. This provides near-continuous flexibility while keeping the calculation process systematic and manageable.
Solution Approach 2:
The patent introduces motion vector difference as an intermediary between the base motion vector and the final refined motion vector. This intermediary layer, controlled by distance and direction syntax elements, allows precise motion vector adjustment without requiring complex direct computation, simplifying the overall system while achieving fine-grained motion representation.
3Measurement precision
If additional motion vector refinement is applied, then prediction accuracy is improved, but computational cost increases
Solution Approach 1:
The patent applies motion vector refinement selectively rather than uniformly across all coding units. By using syntax elements to control when and how motion vector differences are applied, the system achieves high prediction accuracy where needed while avoiding unnecessary computational overhead in cases where simple motion vectors suffice. This partial application strategy balances accuracy and computational cost effectively.
Data Source
AI summary
A method, apparatus, and computer program product provide for improving prediction accuracy of a geometric partitioning merge mode. In the context of a method, the method accesses a coding unit partitioned into a first partition and a second partition. The method also determines whether an enabling value associated with the coding unit is indicative of additional motion vector refinement and, in accordance with a determination that the enabling value associated with the coding unit is indicative of additional motion vector refinement, determines at least one motion vector difference value for at least one partition of the first partition and the second partition. The method additionally applies a prediction of the coding unit in accordance with the at least one motion vector difference value.