CUFED5

我们使用与构建训练数据集相似的方法来收集图像对。总共，测试集包含126组样本。每组分别由一个HR图像和四个级别为L1，L2，L3和L4的参考文献组成。来自测试集的两个示例如图4所示。我们将收集的训练和测试集称为CUFED5，这将在很大程度上促进对RefSR的研究，并为公平比较提供基准。

We collected image pairs using a methodology analogous to that employed for building the training dataset. Overall, the test set encompasses 126 sample groups, with each group comprising one High-Resolution (HR) image and four reference images at levels L1, L2, L3, and L4 respectively. Two examples from the test set are illustrated in Figure 4. We denote the collected training and test datasets as CUFED5, which will significantly advance research on Reference-based Super-Resolution (RefSR) and offer a standardized benchmark for fair comparative assessments.

搜集汇总

背景与挑战

背景概述

CUFED5是一个用于参考图像超分辨率（RefSR）研究的测试数据集，包含126组样本，每组由一个高分辨率图像和四个不同级别的参考文献组成，旨在为RefSR算法提供公平的基准。该数据集由Adobe Research和田纳西大学于2019年发布，属于图像处理领域，常用于评估超分辨率模型的性能。