Motion Vector Difference Signaling for Reference-Aware Inter Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image/video compression technologies are inefficient for high-resolution and high-quality content, particularly in handling motion vector differences, leading to increased transmission and storage costs, and there is a need for improved inter prediction methods to address diverse content types like virtual reality and augmented reality.
Innovation Solution
The method involves efficient signaling of motion vector differences based on reference picture types, allowing for the derivation of L1 motion vector differences, reducing coding complexity and enhancing inter prediction efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of information
If conventional motion vector derivation methods are used for long-term reference pictures, then the motion vector cannot be modified by signaled MVD, but this limits compression efficiency and flexibility
Solution Approach 1:
The patent changes the parameter of motion vector derivation by introducing reference picture type-based constraints. When the reference picture is identified as a long-term reference picture, the motion vector is constrained to use only the motion vector predictor without allowing modification by signaled motion vector difference. This parameter change resolves the contradiction by selectively applying different derivation rules based on reference picture type, improving compression efficiency while managing coding complexity.
Solution Approach 2:
The patent introduces dynamic behavior in motion vector derivation by making the derivation method dependent on the reference picture type. The system dynamically switches between two modes: (1) allowing MVD modification for non-long-term references, and (2) constraining to predictor-only for long-term references. This dynamic adaptation resolves the contradiction by optimizing the derivation process according to the specific reference picture being used.
2Loss of information
If motion vector differences are signaled for all reference pictures, then compression efficiency may improve, but transmission and storage costs increase
Solution Approach 1:
The patent applies local quality by treating long-term reference pictures differently from other reference pictures. Instead of uniformly signaling motion vector differences for all references, the patent locally constrains long-term references to use only motion vector predictors without MVD modification. This selective approach reduces the quantity of signaled data for long-term references while maintaining adequate prediction quality, thus resolving the contradiction between compression efficiency and bitrate.
Solution Approach 2:
The patent applies partial action by signaling motion vector differences only when necessary (for non-long-term reference pictures) and omitting them for long-term references. This partial signaling strategy reduces the overall bitrate while maintaining compression efficiency for the most critical reference pictures, resolving the contradiction between comprehensive signaling and data reduction.
3Stability of the object's composition
If symmetric MVD is applied to all bi-prediction cases, then coding consistency is improved, but complexity increases for diverse reference picture types
Solution Approach 1:
The patent introduces dynamic behavior in MVD application by making it dependent on reference picture type. For long-term reference pictures, symmetric MVD is constrained and not applied, while for other reference pictures, symmetric MVD can be applied normally. This dynamic conditional application maintains coding consistency where appropriate while reducing system complexity by avoiding unnecessary constraints, thus resolving the contradiction.
Solution Approach 2:
The patent applies local quality by differentiating the treatment of motion vectors based on reference picture type. Long-term references receive constrained treatment (no MVD modification) while other references receive full treatment (MVD allowed). This localized differentiation maintains coding consistency for standard cases while simplifying the system by applying special rules only where needed, resolving the contradiction between consistency and complexity.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
According to embodiments of the present document, a prediction procedure may be performed for image/video coding, and the prediction procedure may include symmetric motion vector differences (SMVDs) and merge mode motion vector differences (MMVDs) according to an inter prediction. The inter prediction may be performed on the basis of reference pictures of a current picture, and types (e.g., a long term reference picture, a short term reference picture, etc.) of the reference pictures may be considered for the inter prediction. Therefore, performance and coding efficiency in the prediction procedure can be improved.