Muxer Layer Image Upsampling System Reducing Computation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current image upsampling systems have high computation costs and lack flexibility in adjusting the upsampling multiple, as they rely on standard methods like bicubic and linear upsampling.
Innovation Solution
An image upsampling system utilizing a cascaded structure of convolutional networks and muxer layers, where the muxer layers synthesize feature images to increase resolution by a multiple of n×n times, allowing for flexible adjustment of upsampling multiples and reducing computation through fewer feature images being processed at each stage.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If standard methods like bicubic upsampling and linear upsampling are used, then image resolution can be increased, but computation amount becomes large
Solution Approach 1:
The patent divides the upsampling process into multiple stages, where each stage uses a convolutional network to extract features and a muxer layer to synthesize output. By segmenting the resolution increase into incremental steps (e.g., 2x per stage) rather than applying a single large upsampling operation, the computation amount is reduced while maintaining image quality.
2Manufacturing precision
If standard upsampling methods are used, then image resolution can be increased, but flexibility in adjusting upsampling multiple is limited
Solution Approach 1:
The patent implements a dynamic and flexible upsampling system where the upsampling multiple can be adjusted by configuring the number of cascaded stages and the upsampling factor of each muxer layer. This allows the system to adapt to different resolution requirements without being constrained to fixed upsampling ratios, providing versatility while maintaining computational efficiency.
Data Source
AI summary
An image upsampling system, a training method thereof and an image upsampling method are provided, the feature images of an image are obtained by using the convolutional network, upsampling processing is performed on the images with the muxer layer to synthesize every n×n feature images in the input signal into a feature image with the resolution amplified by n×n times, in the upsampling procedure with the muxer layer, information of respective feature images in the input signal is recorded in the generated feature image(s) without loss; and thus, every time when the image passes through a muxer layer with an upsampling multiple of n, the image resolution can be increased by n×n times.