The Dataset of 32\u00d732 CU Luminance Value Matrices and Corresponding Partition Mode Annotations for the Versatile Video Coding

The dataset is constructed by encoding and decoding YUV-format videos using VTM-23.0 and extracting partition information of 32\u00d732 luma blocks. The dataset includes four folders named after QP values (22, 27, 32, 37). Each folder contains two txt files: the  file with the prefix \lumadata\ stores the luma values of the CU, while the file with the prefix \labels\ stores the partition mode of the CU. The number in each filename indicates the QP value used during encoding.The luminance value matrix of each 32*32CU is stored row by row in the lumadata file:One line of the lumadata file stores the luminance value matrix of a CU, the first row of the matrix occupies bits 0 to 31 of the corresponding row in the lumadata file, the second row occupies positions 32 to 63, and so on. The labels file contains one integer per row, ranging from 0 to 5, each representing a partition pattern: 0 for no split, 1 for vertical binary tree split, 2 for horizontal binary tree split, 3 for vertical ternary tree split, 4 for horizontal ternary tree split, and 5 for quadtree split. Each label corresponds to the partition pattern of the CU stored in the same row of the lumadata file.