Independent Candidate List Generation for Parallel Video Coding
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing video coding technologies face inefficiencies in generating candidate lists for motion information, as they often require sequential processing, which hinders parallel encoding and decoding of motion information across prediction units within a coding unit, leading to reduced throughput.
Innovation Solution
A method where candidate lists for each prediction unit are generated independently without using motion information from other prediction units within the same coding unit, allowing for parallel processing and faster generation of candidate lists and predictive video blocks.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If motion information from other prediction units within the same coding unit is used to generate candidate lists, then the accuracy of motion prediction is improved, but the processing time increases and parallel encoding/decoding is hindered
Solution Approach 1:
The patent divides the candidate list generation process into independent segments for each prediction unit. By generating candidate lists without relying on motion information from other PUs within the same CU, each PU can be processed independently, enabling parallel execution and eliminating sequential processing bottlenecks while maintaining prediction accuracy through alternative candidate selection methods.
Solution Approach 2:
The patent performs preliminary action by pre-generating candidate lists for each prediction unit using only locally available motion information and reference picture data, before any inter-PU dependency would arise. This allows all candidate lists to be prepared in advance and in parallel, ready for subsequent motion compensation operations without causing processing delays.
2Measurement precision
If motion information from other prediction units within the same coding unit is used to generate candidate lists, then the motion vector prediction quality is improved, but the throughput of encoding and decoding is reduced
Solution Approach 1:
The patent segments the candidate list generation into independent units that do not depend on each other. Each prediction unit's candidate list is generated using only its own motion information and reference picture data, allowing simultaneous generation across multiple PUs. This segmentation eliminates the sequential dependency that limits throughput while preserving prediction quality through careful selection of alternative motion candidates.
Solution Approach 2:
The patent uses copying by replicating the candidate list generation process for each prediction unit independently, rather than having each PU access and process motion information from other PUs. This copying approach allows identical processing logic to run in parallel across multiple PUs, increasing throughput while maintaining consistent prediction quality through the same algorithm applied to each unit.
3Measurement precision
If sequential processing is used to generate candidate lists for prediction units, then the motion information accuracy is maintained, but the encoding and decoding efficiency is reduced
Solution Approach 1:
The patent applies segmentation by dividing the candidate list generation into independent, non-dependent segments for each prediction unit. This allows the encoding and decoding process to proceed in parallel across multiple PUs rather than sequentially, dramatically improving efficiency while maintaining motion information accuracy through consistent application of the candidate selection algorithm to each segment.
Solution Approach 2:
The patent performs preliminary action by preparing all candidate lists for all prediction units before the actual motion compensation process begins. Each candidate list is pre-computed using only locally available data, enabling subsequent processing to proceed efficiently in parallel without requiring sequential access to motion information from different PUs, thus maintaining accuracy while boosting overall efficiency.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
For each prediction unit (PU) belonging to a coding unit (CU), a video coder generates a candidate list. The video coder generates the candidate list such that each candidate in the candidate list that is generated based on motion information of at least one other PU is generated without using motion information of any of the PUs belonging to the CU. After generating the candidate list for a PU, the video coder generates a predictive video block for the PU based on one or more reference blocks indicated by motion information of the PU. The motion information of the PU is determinable based on motion information indicated by a selected candidate in the candidate list for the PU.