Adaptive Bi-Prediction Video Coding with Decoder Motion Refinement
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Generalized bi-prediction (GBi) methods in video coding lack support for decoder-side motion derivation tools, such as Decoder-Side Motion Vector Refinement (DMVR) and Pattern-based MV Derivation (PMVD), which are essential for improving coding performance.
Innovation Solution
Extending GBi techniques to support unequal weighting factors for motion vector prediction and derivation, allowing for the use of DMVR and PMVD tools, and adapting weighting factors based on distortion calculations and block sizes to enhance coding efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If GBi methods are used without decoder-side motion derivation tools, then the coding system is simpler, but coding performance is insufficient
Solution Approach 1:
The patent merges GBi methods with decoder-side motion derivation tools (DMVR and PMVD) by integrating them into the GBi framework. The reference blocks from DMVR/PMVD are combined with the GBi prediction process, allowing the system to leverage both the bi-prediction capability and the motion derivation tools simultaneously, thereby improving coding performance without requiring separate independent systems
Solution Approach 2:
The GBi method is extended to serve multiple functions: it not only provides bi-prediction capability but also incorporates decoder-side motion derivation tools (DMVR and PMVD) as part of the same framework. This multi-functionality allows a single GBi-based system to achieve both the benefits of bi-prediction and motion derivation, improving reliability without proportionally increasing system complexity
2Productivity
If equal weighting factors are used for reference blocks, then the processing is simpler, but coding efficiency is suboptimal
Solution Approach 1:
The patent introduces dynamic weighting factors that can be adaptively selected based on distortion calculations between reference blocks and current blocks. Instead of using fixed equal weighting, the system calculates distortions and selects optimal weighting factors from predefined sets, making the weighting mechanism dynamic and adaptive to different coding scenarios, thereby improving coding efficiency
Solution Approach 2:
The patent changes the weighting factor parameter from fixed equal values to variable values that can be selected based on distortion metrics. By introducing a set of possible weighting factors and selecting the optimal one based on calculated distortions, the system optimizes coding efficiency without requiring completely complex parameter estimation, thus balancing productivity improvement with acceptable complexity
3Loss of information
If motion information is explicitly transmitted, then the prediction accuracy is higher, but the bit rate increases
Solution Approach 1:
The patent enables the decoder to derive motion information autonomously using DMVR and PMVD tools without requiring explicit transmission from the encoder. The decoder uses reference blocks and distortion calculations to self-derive the motion vectors, thereby reducing the amount of motion information that needs to be transmitted while maintaining prediction accuracy, thus reducing the quantity of coded bits
4Measurement precision
If decoder-side motion derivation tools are added to GBi, then the prediction accuracy improves, but the processing complexity increases
Solution Approach 1:
The patent segments the motion vector derivation process into distinct components: DMVR for motion vector refinement and PMVD for pattern-based derivation. These segmented functions can be independently applied and combined with GBi, allowing the system to improve prediction accuracy through modular processing rather than requiring a monolithic complex system, thus managing processing complexity
Data Source
AI summary
A method of video coding using generalized bi-prediction (GBi) receives input data associated with a current block in a current picture, wherein the input data comprises information associated with a block size of the current block, determines a set of weighting factor pairs, wherein a size of the set of weighting factor pairs depends on the block size of the current block, and derives a set of advanced motion vector prediction (AMVP) candidate lists comprising MVP (motion vector prediction) candidates. The method further derives a set of final motion information based on the MVP candidates, determines that the set of final information comprises a bi-prediction predictor, generates a final predictor by combining two reference blocks associated with the final motion information using a target weighting factor pair selected from the set of weighting factor pairs, and encodes or decoding the current block using the final predictor.