Selective Attention on Low-Resolution Features for Image Processing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing neural networks for high-quality image processing experience a significant decrease in processing speed due to redundant attention mechanisms that do not contribute to improved image quality, as they are typically applied after generating multiple compressed feature quantities.
Innovation Solution
The proposed information processing apparatus selectively executes attention processing on low-resolution feature quantities, generating high-resolution and low-resolution feature quantities, and combines them to apply predetermined image processing, thereby suppressing processing speed reduction while enhancing image quality.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If attention processing is executed on all compressed feature quantities, then image quality is improved, but processing speed significantly decreases
Solution Approach 1:
The patent applies attention processing selectively only to low-resolution feature quantities rather than all compressed feature quantities. This local application of the attention mechanism focuses computational resources on the most critical feature representations, improving image quality while avoiding the processing speed degradation that would result from applying attention to all feature quantities at all resolution levels.
2Loss of information
If multiple compressed feature quantities are generated and processed through attention mechanisms, then feature representation is enhanced, but redundant processing occurs
Solution Approach 1:
The patent extracts and applies attention processing only to the low-resolution feature quantities, separating this critical processing step from the other compressed feature quantities. By taking out the attention mechanism application to only where it is most needed (low-resolution features), the patent eliminates redundant attention processing on higher-resolution features while preserving the essential feature representation enhancement.
Data Source
AI summary
In order to improve image quality while suppressing a reduction in processing speed, in an information processing apparatus, at least one high-resolution feature quantity is generated for an input image, a low-resolution feature quantity of a lower resolution than the high-resolution feature quantity is generated, attention processing is selectively executed on the low-resolution feature quantity, and the high-resolution feature quantity and the low-resolution feature quantity are combined, and an image to which predetermined image processing has been applied is generated.