Video Decoding Data Dependency Management via Hardware Unit Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing video compression technologies face challenges in limiting data dependencies outside a given block size, which can break parallelism in processing due to tools like Frame Rate Up-Conversion (FRUC) and Local Illumination Compensation (LIC), leading to pipeline stalling.
Innovation Solution
The proposed solution involves decoding and encoding methods that use a flag to determine if a current block can be processed using tools that rely on reconstructed samples from neighboring blocks or larger blocks, ensuring that only blocks with a parent hardware unit are used, thereby restricting data dependencies and allowing independent processing within hardware units.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of energy
If tools like FRUC and LIC are used to improve compression efficiency, then compression efficiency is improved, but data dependencies outside block size increase causing pipeline stalling
Solution Approach 1:
The patent divides the picture into hardware units of a predetermined size and processes each hardware unit independently. By segmenting the processing scope to within hardware unit boundaries, tools like FRUC and LIC can be applied within each unit without creating cross-unit data dependencies, thus maintaining compression efficiency while enabling parallel processing across multiple hardware units.
2Manufacturing precision
If tools using reconstructed samples of neighboring blocks are applied, then decoding accuracy is improved, but processing independence is reduced due to data dependencies
Solution Approach 1:
The patent applies tools like FRUC and LIC locally within each hardware unit using only reconstructed samples from neighboring blocks that are contained within the same hardware unit. This local quality approach ensures that decoding accuracy is improved through the use of these tools while maintaining processing independence, as each hardware unit can be decoded without waiting for other units.
3Device complexity
If block size is increased to reduce the number of processing units, then processing overhead is reduced, but data dependencies increase breaking parallelism
Solution Approach 1:
The patent segments the picture into fixed-size hardware units and restricts tool application within these boundaries. This segmentation allows multiple hardware units to be processed in parallel while keeping the processing overhead manageable through standardized unit sizes. The predetermined hardware unit size optimizes the balance between the number of processing units and the parallel processing capability.
Data Source
AI summary
A decoding method is disclosed. A flag is first decoded. The flag indicates whether a current block of a picture is decoded with a tool of a set of tools. The set of tools comprises tools using reconstructed samples of neighboring blocks decoded prior to the current block or using a block of the picture larger than a hardware unit. The flag is only decoded in the case where the current block has a parent hardware unit. A current block is then decoded responsive to the flag.