Uniform Sub-Block Partitioning for 3D Video View Synthesis Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing three-dimensional video coding systems face inefficiencies in view synthesis prediction due to high complexity and irregular memory access associated with adaptive block partitioning, particularly for large prediction units and non-uniform partitioning.
Innovation Solution
Implementing a method for uniform sub-block partitioning in view synthesis prediction, where a single partition decision is derived based on corner or centric depth samples, allowing for consistent horizontal or vertical splitting of partition blocks into sub-blocks, reducing computational complexity and memory access irregularity.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If adaptive block partitioning is used for VSP, then prediction accuracy is improved, but computational complexity increases significantly
Solution Approach 1:
The patent divides the prediction unit into multiple sub-blocks and applies separate partition decisions to each sub-block rather than the entire PU. This segmentation allows adaptive partitioning to be applied locally where needed while reducing overall computational complexity by avoiding a single complex decision for the entire large block.
Solution Approach 2:
The patent applies adaptive partitioning selectively to sub-blocks rather than uniformly to the entire prediction unit. By using partition flags to indicate whether adaptive partitioning is applied to each sub-block, the system performs partial adaptive action only where beneficial, reducing unnecessary computational overhead while maintaining prediction accuracy where needed.
2Measurement precision
If adaptive block partitioning is used for VSP, then prediction accuracy is improved, but memory access regularity deteriorates
Solution Approach 1:
By segmenting the PU into sub-blocks with independent partition decisions, the patent creates smaller, more manageable memory access patterns. Each sub-block's regular memory access can be processed independently, making it easier to maintain regularity compared to irregular access patterns across a large uniformly partitioned block.
Solution Approach 2:
The patent introduces dynamic partition flags that allow the partitioning structure to adapt at the sub-block level. This dynamic approach enables regular memory access patterns to be maintained within each sub-block while allowing overall flexibility in the partitioning structure, balancing regularity requirements with adaptive partitioning benefits.
3Measurement precision
If multiple partition decisions are made for each 8x8 block, then prediction accuracy is improved, but bitrate for partition flags increases
Solution Approach 1:
The patent applies adaptive partitioning selectively to sub-blocks using partition flags rather than forcing partitioning on every 8x8 block. By using flags to indicate whether adaptive partitioning is applied to each sub-block, the system reduces the total number of partition decisions and associated bitrate overhead while maintaining prediction accuracy where adaptive partitioning provides benefit.
Solution Approach 2:
The patent allows different partitioning strategies to be applied to different sub-blocks based on local characteristics. Each sub-block can have its own partition flag indicating whether adaptive partitioning is applied, allowing the system to optimize for prediction accuracy in regions where it matters while reducing bitrate overhead in regions where simple partitioning suffices.
Data Source
AI summary
A method and apparatus for three-dimensional or multi-view video encoding and decoding using VSP (view synthesis prediction) with uniform sub-block partition are disclosed. For a current texture block comprising multiple partition blocks, the system derives a single partition decision and partitions each partition block of the current texture block into multiples sub-blocks according to the single partition decision. The VSP processing is then applied to each sub-block to derive the inter-view prediction using VSP. The single partition decision is derived using depth samples of the depth block in a reference view.