BIO Coding Unit Partitioning for VPDU-Based Inter Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In the VVC standard, performing inter prediction on PUs larger than VPDUs is challenging due to the difficulty in virtually partitioning PUs into vPUs, requiring a large buffer for gradient and velocity calculations, which increases buffer resource demands and complicates hardware implementation.
Innovation Solution
The processing unit for determining bidirectional prediction is partitioned into units equal to or smaller than the VPDU size, allowing for efficient calculation and determination using partitioned processing units.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If inter prediction is performed on PUs larger than VPDUs using BIO, then encoding efficiency is enhanced, but buffer size increases and hardware implementation becomes complex
Solution Approach 1:
The patent divides a large PU into multiple smaller vPUs (virtual PUs) that can be processed independently. Each vPU is processed through separate gradient and velocity calculation pipelines, allowing parallel processing and reducing the buffer size required for BIO operations while maintaining overall encoding efficiency
Solution Approach 2:
The patent introduces a virtual partitioning dimension by creating vPUs that are logical subdivisions of physical PUs. This allows the system to process large PUs by treating them as collections of smaller virtual units, effectively adding a partitioning dimension to the processing architecture without requiring physically larger buffers
2Measurement precision
If BIO is applied to entire PU, then prediction accuracy improves, but calculation cost increases significantly
Solution Approach 1:
The patent segments the BIO calculation process by applying it to individual vPUs rather than the entire PU. This reduces the calculation cost by limiting the gradient and velocity computations to smaller regions while maintaining prediction accuracy through the cumulative effect of multiple vPU predictions
Solution Approach 2:
The patent applies BIO partially by selecting which vPUs receive full BIO processing based on local characteristics. Some vPUs may receive simplified or skipped BIO processing when full optical flow calculation is not necessary, reducing overall calculation cost while maintaining adequate prediction accuracy
3Measurement precision
If SAD calculation is performed on entire PU to determine BIO application, then determination accuracy improves, but processing time increases
Solution Approach 1:
The patent segments the SAD calculation into per-vPU operations rather than computing it over the entire PU. This allows the system to determine BIO applicability for each vPU independently based on local SAD values, reducing overall processing time while maintaining accurate determination through localized assessments
Data Source
AI summary
The present technology relates to image processing device and method that achieve a reduction in a buffer size.The image processing device partitions a unit of processing into partitioned processing units each of which corresponds to a VPDU size or is equal to or smaller than the VPDU size, the unit of processing being used for calculation of a cost that is used for determining whether or not to perform bidirectional prediction. The image processing device makes the determination by using the cost calculated based on the partitioned processing units. The present technology is applicable to encoding devices or decoding devices.