Geometric Partition Motion Vector Refinement With Adaptive Offset Sets
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing video coding standards, such as VVC and AVS3, face inefficiencies in the geometric partition mode (GPM) due to suboptimal motion vector refinement and signaling overhead, particularly when applied to non-merge inter modes, which limits the coding efficiency of video data compression.
Innovation Solution
Implementing motion vector refinement (MVR) and extending GPM to explicit inter modes by using predefined motion vector difference (MVD) magnitudes and directions, similar to MMVD, to enhance motion accuracy while minimizing signaling overhead.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If motion vector refinement is applied to GPM partitions, then motion accuracy is improved, but signaling overhead increases
Solution Approach 1:
The patent applies parameter changes by using predefined MVR offset sets with different granularity levels (e.g., 1/4-pel, 1/2-pel, 1-pel precision) and allowing adaptive selection between them. The encoder can choose different offset sets based on content characteristics, balancing motion accuracy improvement against signaling overhead for control flags indicating which offset set is used.
Solution Approach 2:
The patent introduces dynamic adaptability by enabling the encoder to switch between different MVR offset sets and refinement modes based on content analysis. The system dynamically adjusts the level of motion vector refinement applied to different GPM partitions, applying finer refinement only where beneficial, thus optimizing the trade-off between accuracy and overhead.
2Productivity
If GPM is extended to explicit inter modes, then coding efficiency is improved, but device complexity increases
Solution Approach 1:
The patent segments the motion vector refinement process into distinct stages: base motion vector derivation, MVR offset selection from predefined sets, and final refined motion vector calculation. This segmentation allows the system to apply complexity only where needed in the encoding pipeline while maintaining efficiency gains from extended GPM support in explicit inter modes.
Solution Approach 2:
The patent employs preliminary action by pre-defining multiple MVR offset sets with different precision levels before encoding begins. These offset sets are prepared in advance and stored in the decoder, eliminating the need for complex real-time calculations during decoding and reducing overall device complexity while maintaining coding efficiency.
3Measurement precision
If multiple MVR offset sets are used, then motion vector precision is improved, but bit rate increases
Solution Approach 1:
The patent applies partial action by allowing the encoder to selectively apply motion vector refinement only to specific GPM partitions where it provides the most benefit, rather than uniformly to all partitions. The encoder can choose to apply MVR to only one partition or both partitions based on content characteristics, reducing the overall bit rate impact while maintaining precision where needed.
Solution Approach 2:
The patent uses copying by reusing the same predefined MVR offset sets across multiple GPM partitions and video sequences. Once the offset sets are defined and transmitted, they can be copied and applied repeatedly without additional signaling overhead, improving motion vector precision without proportionally increasing bit rate.
Data Source
AI summary
Provided is a method for video decoding including: receiving a control variable enabling adaptive switch between motion vector refinement (MVR) offset sets; receiving an indication variable enabling adaptive switch between codeword tables that binarize offset magnitudes in the MVR offset sets under the coding level; partitioning the video block into a first and a second geometric partition; selecting an MVR offset set based on the control variable; receiving syntax elements to determine a first and second MVR offsets applied to the first and second geometric partitions from the selected MVR offset set; obtaining a first and second MVs from a candidate list for the first and the second geometric partition; calculating a first and second refined MVs based on the first and second MVs and the first and second MVR offsets; and obtaining prediction samples based on the first and second refined MVs.