Decoder-Side Motion Vector Refinement for Small Video Blocks
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing video coding technologies face challenges in accurately refining motion vectors, particularly in decoder-side processing, which affects the efficiency and quality of video decoding.
Innovation Solution
The implementation of extended decoder-side motion vector refinement (DMVR) that utilizes bilateral matching cost calculations with different sample sizes for improved motion vector adjustments, allowing for refined motion vectors to be determined and applied to blocks of varying sizes, including subblocks, to enhance decoding accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If decoder side motion refinement (DMVR) is applied to small blocks, then motion vector accuracy is improved, but computational complexity increases significantly
Solution Approach 1:
The patent divides the current block into multiple subblocks and performs DMVR separately on each subblock. This segmentation allows the refinement process to be applied selectively to regions that benefit most from it, reducing overall computational complexity while maintaining accuracy where needed. The patent specifies that DMVR can be applied to subblocks of different sizes (e.g., 4x4, 8x8, 16x16) depending on the block type and prediction mode.
Solution Approach 2:
The patent applies different levels of DMVR processing to different regions of the video data based on local characteristics. Specifically, it performs extended DMVR on certain blocks (e.g., when block size is below a threshold or for specific prediction modes) while using standard DMVR or no refinement for other blocks. This localized approach ensures high motion vector accuracy for challenging regions while avoiding unnecessary computation in simpler regions.
2Measurement precision
If extended DMVR with different sample sizes is used, then decoding accuracy is improved, but processing time increases
Solution Approach 1:
The patent dynamically adjusts the DMVR processing parameters based on block characteristics. It uses extended DMVR (with different sample sizes for cost calculation than reconstruction) selectively for certain block types and sizes, while using standard DMVR for others. The patent specifies conditions such as block size thresholds and prediction mode types that trigger extended DMVR, creating a dynamic processing system that adapts to content requirements.
Solution Approach 2:
The patent changes key parameters of the DMVR process based on block characteristics. It modifies the sample size used for bilateral matching cost calculation independently from the block reconstruction size. The patent introduces parameters such as extended block size for cost calculation (e.g., 2N x 2M for a N x M block) while maintaining the original block size for actual reconstruction, allowing flexibility in optimizing the trade-off between accuracy and processing time.
3Reliability
If DMVR is applied to all blocks, then overall video quality is improved, but computational overhead becomes excessive
Solution Approach 1:
The patent applies partial DMVR processing by selectively enabling extended DMVR for specific block types and sizes rather than all blocks. It uses criteria such as block size thresholds (e.g., applying to blocks smaller than a threshold), prediction mode types (e.g., bi-prediction modes), and block position to determine which blocks receive extended DMVR processing. This partial application maintains video quality while controlling computational overhead.
Data Source
AI summary
The processing circuitry receives a coded video bitstream, the coded video bitstream includes coded information of a first block, the coded information indicates a use of a decoder side motion refinement (DMVR) on the first block. The first block has a first sample size. The processing circuitry applies an extended DMVR on the first block, the extended DMVR uses a second sample size for bilateral matching cost calculation, the second sample size is different from the first sample size of the first block. The processing circuitry calculates a plurality of bilateral matching costs respectively associated with a plurality of motion vector adjustments, at least a first bilateral matching cost in the plurality of bilateral matching costs is calculated based on the second sample size that is different from the first sample size of the first block. The processing circuitry determines a first refined motion vector and reconstructs accordingly.