Refined Motion Vectors for Geometric Video Partitioning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current video coding technologies, such as those in the Versatile Video Coding (VVC) standard, inherit motion vectors for partitions directly from merge candidates without refinement, which may not be accurate, leading to inefficiencies in inter prediction and motion compensation.
Innovation Solution
The proposed solution involves deriving refined motion vectors by adding multiple motion vector differences (MVDs) to a motion vector derived from a merge candidate, allowing for more precise motion compensation and encoding in geometric partitioning modes, such as triangular or geometric merge modes, within video blocks.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If motion vectors are inherited directly from merge candidates without refinement, then device complexity is reduced, but measurement precision of motion compensation deteriorates
Solution Approach 1:
The video block is divided into multiple geometric partitions (triangular or polygonal regions), and each partition is assigned a separate motion vector difference. This segmentation allows different regions to have refined motion vectors tailored to their specific motion characteristics, improving measurement precision without requiring refinement of the entire block at once, thus managing device complexity effectively.
Solution Approach 2:
Motion vector differences are applied locally to specific geometric partitions rather than uniformly across the entire video block. Each partition receives a motion vector difference appropriate to its local motion characteristics, improving the accuracy of motion compensation in regions with varying motion patterns while keeping the overall processing complexity manageable through selective refinement.
2Measurement precision
If multiple motion vector differences are added to refine motion vectors, then measurement precision of motion compensation is improved, but device complexity increases
Solution Approach 1:
The system dynamically determines the number and configuration of geometric partitions based on the content characteristics of the video block. The partitioning strategy and number of motion vector differences applied are adapted to the specific scene, allowing the device complexity to scale with the actual need for precision rather than applying maximum refinement universally.
Solution Approach 2:
The invention changes the parameters of motion vector representation by introducing multiple motion vector differences that are added to the base motion vector. This parameter transformation allows for more precise motion representation through mathematical operations on vector components, improving accuracy while maintaining a systematic approach that can be managed by the processing device.
3Productivity
If geometric partitioning mode is used with refined motion vectors, then video coding efficiency is improved, but device complexity increases
Solution Approach 1:
The video block is divided into geometric partitions (triangular or polygonal regions) that conform to the actual motion patterns in the scene. This segmentation enables more accurate motion compensation by matching partition boundaries with motion discontinuities, improving coding efficiency. The segmentation is performed using standardized geometric modes that manage device complexity through predefined partition templates.
Solution Approach 2:
The geometric partitioning framework provides a universal approach that can handle various motion patterns through a single unified mechanism. The same geometric partitioning infrastructure supports different numbers and configurations of partitions, and can be applied to different video content types, making the increased device complexity worthwhile through broad improvements in video coding efficiency across multiple scenarios.
Data Source
AI summary
Motion vector difference for block with geometric partition is described. One example method of video processing includes determining, for a conversion between a current video block of a video and a bitstream of the current video, that the current video block is coded with a geometric partitioning mode; deriving at least one refined motion vector (MV) for the current video block by adding at least one motion vector difference (MVD) of multiple MVDs signaled or derived for the current video block to a MV derived from a merge candidate associated with the current video block; and performing the conversion based on the refined MV.