Phips/BHI_LR_multi

BHI_LR_multi数据集是BHI SISR数据集的x4低分辨率版本。该数据集通过应用0.25（或x4）缩放比例，并随机使用多种缩放算法（如线性、cubic_mitchell、lanczos、gauss、box和down_up）来创建。这样做的目的是让训练模型能够处理多种缩放情况，而不是仅仅依赖于一种特定的缩放算法。

The BHI_LR_multi dataset is a x4 low-resolution counterpart to the BHI SISR dataset, created by applying multiple scaling algorithms (such as linear, cubic Mitchell, Lanczos, Gaussian, box, and up-down sampling) in a randomized manner. The specific degradation information for each image can be found in the applied_degradations.txt file. The trained model is designed to handle multiple scalings and learn to counteract the effects of these algorithms, such as Lanczos always sharpening and Gaussian always softening. Using this multi-algorithm approach helps the model avoid learning the characteristics of a specific scaling algorithm.