dataset_EOL.mat

This is a subset of training sample, and this subset contains 60,000 training samples and 10,000 test samples. The training sample set is generated based on the images of the ILSVRC2012 validation image set, which contains a total of 50,000 high-quality natural images. For each image (converted into grayscale space at first), Gaussian filtering is used to obtain five versions with different blurred levels. In this paper, the size of the Gaussian filter used is 7*7, and the standard deviation is 2. The source image and the corresponding five blurred versions are given a corresponding image label 6, 5, 4, 3, 2, 1 respectively. When the label value of the first image is greater than the second image, the label values of these sample data are set to 1. Conversely, when the label value of the first image is smaller than the second image, the label values of these sample data are set to 0. For source images and five different blurred versions, the windows with a size of 16 * 16 are randomly select from these images as the sample data. <b></b>