Neural Visual Coding with Dual-Filter Reconstruction Selection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Neural network-based image/video coding quality is expected to be further improved, as existing methods face challenges in adapting to content variations and optimizing rate-distortion performance.
Innovation Solution
A method involving two different filtering processes to generate candidate reconstructions, which are combined to form a target reconstruction, allowing the coding process to adapt to visual data content and enhance coding quality.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If a single filtering process is used to generate reconstruction of visual data component, then device complexity is reduced, but coding quality deteriorates
Solution Approach 1:
The reconstruction process is segmented into multiple independent filtering paths (first filtering process and second filtering process), each generating a candidate reconstruction. This segmentation allows the system to explore multiple reconstruction possibilities without requiring a single complex filter, thereby improving coding quality while managing complexity through modular processing.
Solution Approach 2:
Multiple candidate reconstructions from different filtering processes are merged through selection or combination to produce the final target reconstruction. This merging strategy leverages the strengths of different filtering approaches, achieving superior coding quality by combining rather than relying on a single filtering method.
2Manufacturing precision
If multiple filtering processes are used to generate candidate reconstructions, then coding quality is improved, but device complexity increases
Solution Approach 1:
The system dynamically selects or combines candidate reconstructions from multiple filtering processes based on content characteristics and coding conditions. This dynamic adaptation allows the system to activate only necessary filtering paths for each coding scenario, improving coding quality while preventing unnecessary complexity from being always active.
Solution Approach 2:
Different filtering processes use varying parameters (filter types, strengths, or configurations) to generate diverse candidate reconstructions. By changing parameters across different filtering paths rather than using a single fixed filter configuration, the system achieves better coding quality while maintaining manageable complexity through parameter diversification.
3Adaptability or versatility
If candidate reconstructions are combined to form target reconstruction, then adaptability to content variations is improved, but processing time increases
Solution Approach 1:
Instead of always processing all candidate reconstructions equally, the system applies partial processing by selectively choosing or combining only the most promising candidates based on content analysis. This partial action approach maintains high adaptability to content variations while reducing processing time by avoiding exhaustive processing of all possible reconstructions.
Data Source
AI summary
Embodiments of the present disclosure provide a solution for visual data processing. A method for visual data processing is proposed. The method comprises: determining, for a conversion between visual data and one or more bitstreams of the visual data with a neural network (NN)-based model, a target reconstruction of a first component of the visual data based on a first candidate reconstruction and a second candidate reconstruction of the first component, wherein the first candidate reconstruction is generated based on a first filtering process, and the second candidate reconstruction is generated based on a second filtering process different from the first filtering process; and performing the conversion based on the target reconstruction.