Optical Flow Prediction Refinement for Unified Motion Compensation Precision
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The existing video coding technologies, such as VVC, face inefficiencies in motion compensation due to non-unified precision in optical flow-based prediction refinement tools like BDOF and PROF, leading to difficulties in hardware implementation and suboptimal coding efficiency for affine motion models.
Innovation Solution
Implement bit-depth control for bi-directional optical flow (BDOF) and prediction refinement with optical flow (PROF) by aligning precision of gradient and motion vector calculations to match intermediate prediction sample precision, and harmonizing the workflows to facilitate shared pipeline designs.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If non-unified precision is used in optical flow-based prediction refinement tools, then flexibility in calculation is improved, but hardware implementation complexity increases
Solution Approach 1:
The patent applies parameter changes by standardizing the precision of internal parameters (gradient values, motion difference values) to match the bit depth of prediction samples. This unification of precision parameters simplifies hardware implementation while maintaining calculation flexibility, directly resolving the technical contradiction between adaptability and device complexity.
2Measurement precision
If different precision levels are used for gradient and motion vector calculations, then calculation accuracy is improved, but computational complexity increases
Solution Approach 1:
The patent applies homogeneity by making the precision of all internal parameters uniform and consistent with the prediction sample bit depth. This eliminates the need for multiple precision levels in gradient and motion vector calculations, reducing computational complexity while preserving sufficient accuracy through the unified precision approach.
3Productivity
If separate workflows are used for BDOF and PROF tools, then tool-specific optimization is improved, but hardware pipeline design complexity increases
Solution Approach 1:
The patent applies merging by harmonizing the workflows of BDOF and PROF tools to use identical calculation procedures and parameter precisions. This allows the hardware pipeline to share common processing stages for both tools, reducing pipeline design complexity while maintaining the ability to independently optimize each tool's performance through selective activation.
Data Source
AI summary
A method for video encoding is provided. The method includes determining, by an encoder, a reference picture associated with a video block; determining, by the encoder, initial prediction samples of the video block from the reference picture; deriving, by the encoder, internal prediction refinement with optical flow (PROF) parameters of a PROF derivation process by applying right-shifting operations, wherein the internal PROF parameters comprise horizontal gradient values, vertical gradient values, horizontal motion difference values, and vertical motion difference values derived for samples in the video block; determining, by the encoder, prediction refinement values for the samples in the video block based on the horizontal gradient values, the vertical gradient values, the horizontal motion difference values, and the vertical motion difference values; and determining, by the encoder, refined prediction samples of the video block based on the initial prediction samples and the prediction refinement values.