Video Coding Motion Vector Prediction Using Merge Modes
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing video coding technologies face challenges in efficiently reducing redundancy in video data, particularly in inter-picture prediction, where motion vectors (MVs) are not optimally predicted, leading to suboptimal compression ratios and increased data requirements.
Innovation Solution
The proposed solution introduces enhanced MV prediction mechanisms, including MMVD, sub-block affine merge prediction, and multi-hypothesis intra prediction, to improve the efficiency of motion vector coding by deriving MVs from refined merge modes and spatial or temporal neighbors, reducing the need for direct MV coding.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If motion vectors are coded directly without prediction, then coding simplicity is maintained, but compression efficiency deteriorates due to high bitrate requirements
Solution Approach 1:
The patent applies preliminary action by predicting motion vectors from neighboring blocks before actual coding. The merge mode and AMVP mechanism pre-calculate candidate motion vectors from spatially adjacent blocks, so that the current block's motion vector can be derived with minimal additional data, improving compression efficiency while managing complexity through structured prediction workflows
Solution Approach 2:
The patent uses intermediary elements in the form of motion vector candidates and candidate lists. Instead of directly coding motion vectors, the system introduces intermediate prediction candidates from neighboring blocks (A0, A1, B0, B1, etc.) that serve as mediators to derive the final motion vector, thereby reducing the bitrate required for direct motion vector transmission
2Productivity
If motion vectors are predicted from neighboring blocks, then compression ratio is improved, but prediction accuracy may deteriorate in complex motion scenarios
Solution Approach 1:
The patent applies segmentation by dividing the motion vector prediction into multiple candidate blocks (A0, A1, B0, B1, B2) surrounding the current block. Each neighboring block provides a separate prediction candidate, allowing the encoder to select the most accurate one or combine multiple candidates, thereby maintaining compression efficiency while improving motion vector accuracy in complex scenarios
Solution Approach 2:
The patent uses composite prediction by combining motion vector information from multiple neighboring blocks. The AMVP mode and merge mode synthesize candidates from different spatial locations (above, below, left, right neighbors), creating a composite prediction that is more robust to complex motion patterns than single-block prediction alone
3Productivity
If multiple prediction modes are implemented, then compression efficiency is enhanced, but computational complexity increases
Solution Approach 1:
The patent applies dynamics by implementing multiple prediction modes (merge mode, AMVP mode, refined merge mode) that can be dynamically selected based on the characteristics of each block. The encoder adapts between different prediction strategies depending on motion complexity, neighboring block availability, and rate-distortion optimization results, enhancing compression efficiency while managing computational complexity through selective mode application
Data Source
Figure 1
Figure 2
Figure 3
AI summary
Aspects of the disclosure provide methods and apparatuses for video coding. In some examples, an apparatus includes processing circuitry that obtains a first syntax element from a coded video bitstream, the first syntax element being associated with a plurality of modes for identifying and applying motion information to a current block, and a first value of the first syntax element indicating a subset of the plurality of modes for the current block. The processing circuitry further generates a predictor of the current block according to the subset of the plurality of modes indicated by the first value of the first syntax element, and generates reconstructed samples of the current block based on the predictor.