Your Super Resolution Model is not Enough for Tackling Real-World Scenarios
Dongsik Yoon, Jongeun Kim
TL;DR
This work tackles the limitation of fixed-scale SR models in real-world contexts by introducing SAAM, a lightweight plug-in that enables arbitrary-scale upsampling when attached to modern SR backbones. SAAM uses a SimAM-guided hourglass to generate a scale-conditioned guidance map and a lightweight scale-aware convolution to produce adaptive features, fused via $F' = F + F_{ ext{adpt}} \times M$, together with a gradient-variance loss to sharpen edges. The approach is validated across multiple backbones (SCNet, HiT-SR, OverNet) and datasets, showing competitive or superior performance for both integer and non-integer scales with minimal parameter overhead. The results demonstrate SAAM’s practicality for real-world SR tasks, including mobile and edge deployments, by achieving robust multi-scale upsampling without significant computational burden.
Abstract
Despite remarkable progress in Single Image Super-Resolution (SISR), traditional models often struggle to generalize across varying scale factors, limiting their real-world applicability. To address this, we propose a plug-in Scale-Aware Attention Module (SAAM) designed to retrofit modern fixed-scale SR models with the ability to perform arbitrary-scale SR. SAAM employs lightweight, scale-adaptive feature extraction and upsampling, incorporating the Simple parameter-free Attention Module (SimAM) for efficient guidance and gradient variance loss to enhance sharpness in image details. Our method integrates seamlessly into multiple state-of-the-art SR backbones (e.g., SCNet, HiT-SR, OverNet), delivering competitive or superior performance across a wide range of integer and non-integer scale factors. Extensive experiments on benchmark datasets demonstrate that our approach enables robust multi-scale upscaling with minimal computational overhead, offering a practical solution for real-world scenarios.