Decoder-Side Motion Vector Refinement Block Partitioning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current video coding techniques, such as those used in the HEVC standard, face challenges in efficiently managing decoder-side motion vector refinement (DMVR), which increases computation workload and processing time without justifying the improvement in video quality.
Innovation Solution
The method involves early termination or disabling of DMVR when the video quality improvement does not justify the increased processing time, by selectively terminating motion vector refinement based on performance measures such as similarity and distortion between reference blocks, and partitioning larger blocks into smaller sub-blocks for independent processing.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If decoder-side motion vector refinement (DMVR) is applied to all blocks, then motion vector accuracy is improved, but computation workload and processing time increase significantly
Solution Approach 1:
The patent divides the video picture into different block types and applies DMVR selectively only to certain blocks (e.g., blocks with specific motion characteristics or block sizes), rather than uniformly applying it to all blocks. This segmentation approach maintains motion vector accuracy for blocks that benefit from refinement while avoiding unnecessary computation for blocks where refinement provides minimal improvement.
Solution Approach 2:
The patent implements early termination of DMVR refinement when a predetermined number of iterations is reached or when a convergence criterion is met. Instead of performing exhaustive refinement on all blocks, the system performs partial refinement only where and when needed, thereby reducing overall computation workload while maintaining sufficient motion vector accuracy for video quality.
2Productivity
If DMVR is applied to large blocks, then coding efficiency is improved, but processing time increases due to the size of the block
Solution Approach 1:
The patent applies early termination criteria to DMVR processing of large blocks, limiting the refinement iterations based on block size. For larger blocks, the system performs fewer refinement iterations or applies coarser refinement steps, thereby maintaining coding efficiency improvements from DMVR while reducing the processing time that would otherwise scale with block size.
3Measurement precision
If motion vector refinement iterations are increased, then motion vector precision is improved, but computation workload increases
Solution Approach 1:
The patent implements early termination of the motion vector refinement process by limiting the number of iterations to a predetermined value or by monitoring convergence criteria. The system performs refinement iterations only until a satisfactory precision level is achieved or the iteration limit is reached, avoiding excessive computation that would occur with unlimited iterations. This partial action approach maintains sufficient motion vector precision while significantly reducing computation workload.
Solution Approach 2:
The patent employs adaptive refinement strategies where the number of refinement iterations or the refinement step size is dynamically adjusted based on block characteristics, motion complexity, or intermediate refinement results. For simple blocks or blocks already close to optimal motion vectors, the system performs fewer iterations or coarser refinement, while allocating more computational resources to complex blocks that benefit from higher precision. This dynamic approach optimizes the balance between motion vector precision and computation workload.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
Aspects of the disclosure provide a method and an apparatus for video coding. A processing circuitry determines a process unit size for a decoder-side motion vector refinement (DMVR). Then, the processing circuitry enables an application of the DMVR on a current block when a size of the current block is smaller than or equal to the process unit size; and disables the application of the DMVR on the current block when the size of the current block is larger than the process unit size. In an example, the application of the DMVR refines the motion vectors with or without a template block.