Efficient Cost-and-Quality Controllable Arbitrary-scale Super-resolution with Fourier Constraints

2025-10-27

AI Summary

Existing arbitrary-scale super-resolution methods typically employ RNNs to predict Fourier coefficients sequentially, leading to error accumulation, inefficient inference, and difficulty in balancing reconstruction quality against computational cost. To address these limitations, this paper proposes the first multi-Fourier-component joint prediction framework for multi-scale super-resolution. Our approach introduces Fourier-domain unified constraint modeling, casting spectral coefficient prediction as an end-to-end differentiable optimization problem. By integrating differentiable upsampling with frequency-domain attention, the model generates high-fidelity reconstructions at arbitrary scales in a single forward pass. Crucially, it enables fine-grained, continuous trade-offs between computational cost and reconstruction quality. Extensive experiments on multiple benchmarks demonstrate that our method achieves significant PSNR and SSIM improvements over prior arts while requiring substantially fewer FLOPs—delivering efficient, controllable, and high-quality super-resolution.

Technology Category

• Computer Vision: Multi-modal Vision
• Machine Learning: Multimodal Learning
• Search and Optimization: Mixed Discrete/Continuous Search

Application Category

• Search and Retrieval-Augmented AI: Efficiency and scalability of Web search engines
• Graph Algorithms and Modeling for the Web: Algorithms and analysis for heterogeneous, signed, attributed, multi-relational, temporal, higher-order, and annotated Web-related graphs
• User Modeling, Personalization and Recommendation: Federated recommendation systems and personalization

Abstract

Cost-and-Quality (CQ) controllability in arbitrary-scale super-resolution is crucial. Existing methods predict Fourier components one by one using a recurrent neural network. However, this approach leads to performance degradation and inefficiency due to independent prediction. This paper proposes predicting multiple components jointly to improve both quality and efficiency.

Problem

• Enhancing arbitrary-scale super-resolution efficiency and quality
• Overcoming performance degradation from independent Fourier prediction
• Jointly predicting multiple Fourier components for better results

Innovation

• Jointly predicts multiple Fourier components
• Improves super-resolution quality and efficiency
• Enables cost-and-quality controllable arbitrary scaling