Disabling Decoder Motion Vector Derivation for Video Coding
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The combination of decoder side motion vector refinement techniques like DMVR and multiple hypothesis prediction (MHP) in video coding increases computational complexity significantly, affecting implementation efficiency without substantial improvements in coding performance.
Innovation Solution
Disabling decoder side motion vector derivation techniques such as DMVR or bi-directional optical flow (BDOF) when a merge candidate includes an additional inter prediction signal, to reduce implementation complexity while maintaining coding performance.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If decoder side motion vector derivation techniques (DMVR, BDOF) are combined with multiple hypothesis prediction (MHP), then coding performance is improved, but computational complexity increases significantly
Solution Approach 1:
The patent extracts and separates the processing paths by disabling DMVR and BDOF when MHP is detected. The decoder checks for the presence of additional inter prediction signals and conditionally disables complex motion vector derivation techniques, thereby removing the harmful computational complexity while preserving coding performance through the alternative MHP approach.
Solution Approach 2:
Instead of the conventional approach of applying DMVR/BDOF to refine motion vectors and then applying MHP, the patent inverts the logic by detecting MHP first and preventing DMVR/BDOF application. This reversal avoids the redundant and computationally expensive combination while maintaining the benefits of multiple hypothesis prediction.
2Measurement precision
If decoder side motion vector derivation techniques are applied to base prediction signal, then motion vector accuracy is improved, but implementation complexity increases
Solution Approach 1:
The patent introduces dynamic conditional logic that adjusts the processing pipeline based on the presence of additional inter prediction signals. The decoder dynamically disables DMVR and BDOF operations when MHP is detected, creating an adaptive system that optimizes the balance between motion vector accuracy and implementation complexity based on the specific coding scenario.
Solution Approach 2:
The patent sacrifices the potentially beneficial but computationally expensive DMVR and BDOF operations in favor of the more efficient MHP approach. By conditionally disabling these complex techniques when additional inter prediction signals are present, the system uses simpler, more efficient processing paths that achieve comparable or sufficient coding performance without the overhead of multiple derivation techniques.
3Measurement precision
If multiple inter prediction signals are used, then prediction accuracy is improved, but processing complexity increases
Solution Approach 1:
The patent extracts and removes the redundant DMVR and BDOF processing steps from the prediction pipeline when additional inter prediction signals are present. By detecting the presence of multiple hypotheses and conditionally disabling complex motion vector derivation, the system keeps only the necessary processing operations, thereby reducing overall complexity while maintaining the accuracy benefits of multiple prediction signals.
Data Source
AI summary
A method of decoding video data may comprise determining merge candidates for a block of video data and determining if a merge candidate of the merge candidates includes an additional inter prediction signal. If a merge candidate includes an additional inter prediction signal, the method may include disabling at least one decoder side motion vector derivation technique for use on a base prediction signal of the block of video data.