Local Implicit Normalizing Flow for Arbitrary-Scale Image Super-Resolution
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing arbitrary-scale image super-resolution methods produce blurry outputs due to ignoring the ill-posed problem and using per-pixel absolute loss, limiting their flexibility and quality in real-world applications.
Innovation Solution
The Local Implicit Normalizing Flow (LINF) framework models the distribution of texture details under different scaling factors using a coordinate conditional normalizing flow and local implicit module, enabling the generation of photorealistic high-resolution images by learning the distribution of local texture patches and controlling sampling temperature for improved fidelity and perceptual quality.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If arbitrary-scale SR methods use per-pixel absolute loss to train the model, then the model can perform arbitrary-scale super-resolution, but the output images become blurry
Solution Approach 1:
The patent changes the loss function parameter from per-pixel absolute loss to perceptual loss (LPIPS) and incorporates style loss, transforming the optimization objective to prioritize perceptual quality and texture fidelity over pixel-level accuracy, thereby producing sharp images with rich details while maintaining arbitrary-scale capability
Solution Approach 2:
The patent introduces perceptual loss feedback mechanisms that evaluate image quality through learned feature representations rather than direct pixel comparison, enabling the model to iteratively improve texture synthesis and sharpness while maintaining the desired scale flexibility
2Reliability
If flow-based methods learn the distribution of high-resolution images with normalizing flow, then the method can address the ill-posed nature of super-resolution, but the method is limited to predefined fixed-scale SR
Solution Approach 1:
The patent makes the normalizing flow model dynamic by introducing scale-conditioned transformations where the flow parameters are modulated by the target scale factor, allowing the same model architecture to adaptively handle arbitrary scaling operations while maintaining reliable ill-posed problem solving through distribution learning
Solution Approach 2:
The patent creates a universal super-resolution model that can perform multiple scaling operations (2x, 4x, 8x, and arbitrary scales) using a single trained normalizing flow framework, eliminating the need for separate fixed-scale models while maintaining reliable reconstruction quality across all scales
Data Source
AI summary
A method of local implicit normalizing flow for arbitrary-scale image super-resolution, an associated apparatus and an associated computer-readable medium are provided. The method applicable to a processing circuit may include: utilizing the processing circuit to run a local implicit normalizing flow framework to start performing arbitrary-scale image super-resolution with a trained model of the local implicit normalizing flow framework according to at least one input image, for generating at least one output image, where a selected scale of the output image with respect to the input image is an arbitrary-scale; and during performing the arbitrary-scale image super-resolution with the trained model, performing prediction processing to obtain multiple super-resolution predictions for different locations of a predetermined space in a situation where a same non-super-resolution input image among the at least one input image is given, in order to generate the at least one output image.