Video Decoding Motion Vector Refinement for Low-Complexity Bi-Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Bi-directional prediction in video coding schemes, such as DMVR processing, results in high complexity due to the modification of motion vectors using two prediction images, which is computationally intensive and inefficient.
Innovation Solution
An image decoding and coding apparatus that modifies motion vectors only when neither prediction image is weighted, and includes a bi-directional optical flow sample prediction processing unit that determines the application of processing based on error values, reducing overall processing complexity.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If DMVR processing and BDOF processing are applied to achieve high image quality, then prediction image quality is improved, but processing complexity increases
Solution Approach 1:
The patent changes the parameter of motion vector modification by introducing conditional logic: DMVR processing is applied only when weighted prediction is not used. This parameter change (conditional application based on prediction type) reduces processing complexity while maintaining prediction image quality for non-weighted cases.
Solution Approach 2:
The patent applies DMVR processing partially rather than universally - specifically, only for non-weighted prediction cases. This partial action approach avoids unnecessary processing in weighted prediction scenarios, thereby reducing overall complexity while maintaining high image quality where it matters most.
2Measurement precision
If motion vectors are modified using error evaluation of two prediction images, then prediction accuracy is improved, but processing amount increases
Solution Approach 1:
The patent changes the processing parameter by conditionally applying error evaluation only to non-weighted prediction cases. This parameter modification reduces the quantity of processing operations while maintaining motion vector accuracy for the applicable cases.
Solution Approach 2:
The patent segments the prediction processing into two distinct paths: weighted prediction (without DMVR) and non-weighted prediction (with DMVR). This segmentation allows error evaluation to be applied only where beneficial, reducing overall processing amount while maintaining accuracy where needed.
3Manufacturing precision
If bi-directional optical flow sample prediction processing is always applied, then prediction image quality is improved, but processing complexity and time increase
Solution Approach 1:
The patent applies bi-directional optical flow sample prediction processing partially - only when weighted prediction is not used. This partial application maintains high prediction image quality for non-weighted cases while improving processing efficiency by avoiding unnecessary computations in weighted prediction scenarios.
Solution Approach 2:
The patent changes the processing parameter by making bi-directional optical flow sample prediction conditional on the prediction type. This parameter change (conditional application) reduces processing time and complexity while maintaining high image quality where the processing is applicable.
Data Source
AI summary
An image decoding apparatus and an image coding apparatus that reduce complexity of high image quality processing are implemented. The image decoding apparatus includes an inter prediction parameter decoding processing unit that includes processing of, regarding two motion vectors, modifying the two motion vectors from an error of two prediction images. In a case that neither of the two prediction images is a case of weighted prediction, the processing of modifying the two motion vectors is performed.