Neural Compression Precision Allocation for Efficient Visual Data Conversion
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing neural network-based image and video compression methods face challenges in precision setting, leading to inefficiencies in encoding and decoding times, and are limited by device capabilities.
Innovation Solution
A method for visual data processing that determines precision information for multiple modules based on neural network models, allowing for proper precision levels to be applied during conversion, thereby improving coding efficiency and effectiveness.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If fixed high precision is used for all modules in neural network-based compression, then measurement precision is improved, but device complexity and loss of time increase due to unnecessary computational overhead
Solution Approach 1:
The patent applies dynamic precision adjustment by determining appropriate precision levels for different modules based on their specific requirements and the characteristics of the visual data being processed. This allows the system to adapt precision dynamically rather than using a fixed high precision for all modules, thereby reducing unnecessary computational complexity while maintaining required measurement precision.
Solution Approach 2:
The patent implements local quality by assigning different precision levels to different modules according to their specific needs. Instead of uniformly applying high precision across all modules, each module receives the appropriate precision level required for its function, optimizing the balance between measurement precision and device complexity.
2Manufacturing precision
If high precision is used for all modules, then manufacturing precision is improved, but productivity decreases due to increased encoding and decoding times
Solution Approach 1:
The system dynamically determines appropriate precision levels for each module based on the visual data characteristics and module requirements. This dynamic approach maintains high coding precision where needed while reducing precision for modules where it is not necessary, thereby improving encoding speed and overall productivity without sacrificing required coding precision.
Solution Approach 2:
The patent changes the precision parameter adaptively for different modules and processing stages. By adjusting the precision parameter based on specific requirements rather than maintaining a fixed high value throughout, the system achieves both high coding precision where necessary and improved productivity through reduced computational burden in other areas.
3Manufacturing precision
If uniform precision is applied across all modules, then manufacturing precision is maintained, but adaptability decreases due to inability to optimize for different data characteristics
Solution Approach 1:
The patent implements dynamic precision determination that adapts to different visual data characteristics and module requirements. This allows the system to maintain required coding precision while simultaneously adapting to various data types, resolutions, and content characteristics, thereby improving both manufacturing precision and adaptability.
Solution Approach 2:
The system applies local quality by determining appropriate precision levels for each module based on its specific function and the local characteristics of the visual data being processed. This localized precision optimization maintains high coding precision where needed while providing adaptability to different data characteristics across different modules.
Data Source
AI summary
Embodiments of the present disclosure provide a solution for visual data processing. A method for visual data processing comprises: determining, for a conversion between a current visual unit of visual data and a bitstream of the visual data, precision information indicating at least one precise level for a plurality of modules, at least one of the plurality of modules being based on a neural network model; and performing the conversion by applying the plurality of modules to the current visual unit based on the precision information.