Video Decoding Merge Candidate List Parallelization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional motion compensation methods in video encoding/decoding are limited by dependency between merge candidate derivation processes, leading to reduced encoding/decoding efficiency and increased hardware complexity, particularly due to the use of only uni-directional and bi-directional predictions and complex memory access patterns.
Innovation Solution
The method involves generating a merge candidate list that includes spatial, temporal, and combined merge candidates, allowing for uni-directional, bi-directional, tri-directional, and quad-directional predictions, and performing these processes in parallel to simplify hardware logic and increase throughput.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If conventional merge mode with only spatial, temporal, and bi-prediction merge candidates is used, then device complexity is reduced, but encoding efficiency is limited
Solution Approach 1:
The patent combines spatial merge candidates, temporal merge candidates, and new combined merge candidates into a unified merge candidate list. The combined merge candidate is derived by combining motion information from both spatial and temporal references, enabling more versatile prediction modes (tri-directional and quad-directional predictions) while maintaining systematic organization, thus improving encoding efficiency without excessively increasing device complexity
Solution Approach 2:
The patent introduces dynamic derivation of combined merge candidates based on availability conditions of spatial and temporal merge candidates. The merge candidate list is dynamically constructed by checking availability flags and selectively adding candidates, allowing the system to adapt to different coding scenarios and maximize encoding efficiency while managing complexity through conditional logic
2Reliability
If dependency between temporal merge candidate derivation and bi-prediction merge candidate derivation is maintained, then processing accuracy is ensured, but throughput is limited
Solution Approach 1:
The patent segments the merge candidate derivation process into independent parallel streams: spatial merge candidate derivation, temporal merge candidate derivation, and combined merge candidate derivation. Each stream operates independently with its own availability checks and derivation logic, removing interdependencies and enabling parallel processing across multiple pipelines, thus increasing throughput while maintaining derivation accuracy through independent validation of each candidate type
3Measurement precision
If bi-prediction merge candidate derivation is performed, then prediction accuracy is improved, but memory access bandwidth increases
Solution Approach 1:
The patent applies local quality optimization by deriving combined merge candidates only when both spatial and temporal merge candidates are available and suitable for the current block. The availability checks and conditional derivation ensure that memory access is performed only when necessary, reducing unnecessary bandwidth consumption while maintaining high prediction accuracy where combined candidates provide benefit
4Manufacturing precision
If zero merge candidate derivation is performed differently according to slice type, then coding precision is improved, but hardware logic becomes complex
Solution Approach 1:
The patent implements a universal zero merge candidate derivation mechanism that handles all slice types through a unified logic structure. The same derivation process and availability checks are applied regardless of slice type, eliminating the need for separate derivation paths for different slice types. This universal approach maintains coding precision by consistently applying the same rules while significantly simplifying hardware logic by removing type-specific branching
Data Source
AI summary
The present invention relates to a method for encoding/decoding a video. To this end, the method for decoding a video may include: generating a merge candidate list of a current block including at least one merge candidate corresponding to each of a plurality of reference picture lists; determining at least one piece of motion information by using the merge candidate list; and generating a prediction block of the current block by using the determined at least one piece of motion information.