Rethinking Imbalance in Image Super-Resolution for Efficient Inference

Abstract

Existing super-resolution (SR) methods optimize all model weights equally using L 1 or L 2 losses by uniformly sampling image patches without considering dataset imbalances or parameter redundancy, which limits their performance. To address this issue, we formulate the image SR task as an imbalanced distribution transfer learning problem from a statistical probability perspective and propose a plug-and-play Weight-Balancing framework (WBSR) for image SR to achieve balanced model learning without changing the original model structure or training data. Specifically, we develop a Hierarchical Equalization Sampling (HES) strategy to address data distribution imbalances, enabling better feature representation from texture-rich samples. To tackle model optimization imbalances, we propose a Balanced Diversity Loss (BDLoss) function, focusing on learning texture regions while disregarding redundant computations in smooth regions. After joint training of HES and BDLoss to rectify these imbalances, we present a gradient projection dynamic inference strategy to facilitate accurate and efficient reconstruction during inference. Extensive experiments across various models, datasets, and scale factors demonstrate that our method achieves comparable or superior performance to existing approaches with approximately a 34% reduction in computational cost. The code is available at https://github.com/aipixel/WBSR .