Triangular Prediction Unit Mode for Video Coding Efficiency
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
High-Efficiency Video Coding (HEVC) faces challenges in achieving optimal coding efficiency due to limitations in prediction modes, particularly in Skip and Merge modes, where motion inference methods are used but lack efficient handling of block partitions and motion vector prediction.
Innovation Solution
The implementation of a target merge mode using triangular prediction unit mode (TPM), which splits coding units into triangular prediction units with adaptive weighting and motion vector storage settings, enabling efficient prediction and motion information storage across block partitions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If Skip and Merge modes utilize motion inference methods from spatially neighboring blocks or temporal blocks, then the coding complexity is reduced and fewer motion information parameters need to be transmitted, but the coding efficiency is insufficient due to lack of efficient handling of block partitions and motion vector prediction
Solution Approach 1:
The current block is divided into two triangular prediction units (first and second prediction units) by a partitioning line. Each triangular prediction unit independently derives motion vector predictors from its own candidate set, allowing separate optimization of motion compensation for different regions while maintaining overall coding efficiency
Solution Approach 2:
Different motion vector prediction strategies are applied to different triangular prediction units based on their local characteristics. Each unit selects from its specific candidate set (spatial candidates, temporal candidates, or combined candidates) to match local motion patterns, improving prediction accuracy without uniformly increasing complexity across the entire block
2Measurement precision
If triangular prediction unit mode splits coding units into triangular prediction units with adaptive weighting, then the prediction accuracy is improved for different motion regions, but the computational complexity increases due to multiple motion information derivations
Solution Approach 1:
The method applies triangular partitioning and multiple motion information derivations selectively based on block characteristics (e.g., when motion discontinuities are detected or block size exceeds thresholds). Not all blocks undergo full triangular prediction processing, balancing accuracy improvement with computational cost by applying the complex operation only where beneficial
3Measurement precision
If multiple motion information derivations are performed for triangular prediction units and overlap prediction regions, then the motion compensation accuracy is enhanced, but the number of parameters to be encoded and transmitted increases
Solution Approach 1:
Motion information from multiple sources (spatial candidates, temporal candidates, and combined candidates) is merged to form a unified motion vector predictor for each triangular prediction unit. The candidate sets from different sources are combined and the best predictor is selected, reducing the need to transmit multiple separate motion parameters while maintaining high prediction accuracy
Data Source
AI summary
A video coder receives data from a bitstream for a block of pixels to be encoded or decoded as a current block of a current picture of a video. Upon determining that an applied block setting of the current block satisfies a threshold condition, the video coder generates a first prediction based on a first motion information for a first prediction unit of the current block. The video coder generates a second prediction based on a second motion information for a second prediction unit of the current block. The video coder generates a third prediction based on the first and second motion information for an overlap prediction region that is defined based on a partitioning between the first prediction unit and the second prediction unit. The video coder encodes or decodes the current block by using the first, second, and third predictions.