DIV2K_train_BLUR_90_gray

We cropped the original DIV2K validation images into non-overlapping 90 \u00d7 90 pixel patches and converted them to single-channel grayscale. Each sharp patch was then convolved with a single 9 \u00d7 9 isotropic Gaussian kernel (\u03c3 = 1.0) . The set is intended for benchmarking grayscale blind deblurring, kernel-estimation networks, or as a lightweight validation subset for larger super-resolution\/deblurring pipelines.