Adaptive Bit-Width Control for PROF Video Prediction Refinement
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The existing video coding standards, such as VVC, HEVC, and AVC, face inefficiencies in motion compensation due to limitations in bit-depth representation and precision in prediction refinement tools like PROF and BDOF, leading to suboptimal coding efficiency and hardware implementation challenges.
Innovation Solution
The proposed solution involves controlling internal bit-widths of the PROF derivation process by applying right-shifting to internal parameters, harmonizing the bit-depth representation of gradients and motion differences, and aligning the workflows of PROF and BDOF to facilitate hardware implementations, while adaptively disabling PROF for affine coding blocks with small motion differences.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If high precision internal PROF parameters are used, then prediction refinement accuracy is improved, but computational complexity and hardware implementation difficulty increase
Solution Approach 1:
The patent applies parameter changes by dynamically adjusting the bit-width of internal PROF parameters based on motion difference magnitude. When motion differences are small, lower precision (narrower bit-width) is used; when motion differences are large, higher precision (wider bit-width) is applied. This adaptive parameter adjustment resolves the contradiction by matching precision requirements to actual motion characteristics, reducing unnecessary computational complexity while maintaining prediction accuracy where needed.
Solution Approach 2:
The patent introduces dynamic bit-width adjustment for internal PROF parameters based on motion difference thresholds. The system transitions between different precision levels dynamically during the prediction refinement process, rather than using fixed high precision throughout. This dynamic approach reduces average computational complexity while preserving accuracy for significant motion cases.
2Measurement precision
If PROF is applied to all affine coding blocks, then prediction quality is improved, but computational complexity increases
Solution Approach 1:
The patent applies partial action by selectively disabling PROF for affine coding blocks with small motion differences. Instead of applying PROF uniformly to all affine blocks, the system identifies blocks where motion differences fall below a threshold and skips the refinement process for those blocks. This partial application of PROF maintains prediction quality for blocks that need it while improving overall coding efficiency by avoiding unnecessary computations on blocks with minimal motion variation.
Solution Approach 2:
The patent applies local quality by treating different affine coding blocks differently based on their motion characteristics. Blocks with large motion differences receive full PROF treatment for high prediction quality, while blocks with small motion differences receive reduced or no PROF treatment. This localized quality adjustment optimizes the balance between prediction quality and computational complexity for each block individually.
3Measurement precision
If different bit-depth representations are used for gradients and motion differences, then representation precision is improved, but device complexity increases
Solution Approach 1:
The patent applies parameter changes by harmonizing the bit-depth representations of gradients and motion differences to match the internal PROF parameter precision. This alignment ensures that arithmetic operations between these parameters maintain consistent precision levels, avoiding the need for complex mixed-precision arithmetic logic. The parameter harmonization reduces processing complexity while preserving adequate representation precision for both gradients and motion differences.
Data Source
AI summary
A bit-width representation method of prediction refinement with optical flow (PROF), apparatus, and a non-transitory computer-readable storage medium are provided. The method includes obtaining a reference picture I associated with a video block within the video signal, obtaining prediction samples I(i,j) of the video block from a reference block in the reference picture I, controlling internal bit-widths of a PROF derivation process for various representation precisions of internal PROF parameters by applying right-shifting to the internal PROF parameters based on different bit-shift values, obtaining prediction refinement values for samples in the video block based on the PROF derivation process being applied to the video block based on the prediction samples I(i,j), and obtaining prediction samples of the video block based on the combination of the prediction samples and the prediction refinement values.