A Super Resolution Dataset of Digital Rocks (DRSRD1): Sandstone and Carbonate

Description

Digital Rocks Super Resolution Dataset 1: Sandstone and Carbonate micro-CT Images used for 2D and 3D Super Resolution Machine Learning Training, Validation, and Testing. This dataset comprises of organised 2D slices and 3D volumes of Bentheimer Sandstone [sourced internally from UNSW] and Estaillades Carbonate [https://www.digitalrocksportal.org/projects/58/origin_data/209/], imaged at a resolution of 3.8 and 3.1 microns respectively. The dataset is organised in a similar fashion to the DIV2K dataset : (https://data.vision.ee.ethz.ch/cvl/DIV2K/). To the knowledge of the authors, this dataset is the first of its kind to be used for the purpose of benchmarking Super Resolution algorithms in digital rock images. This dataset has been used thus far to test several Super Resolution Convolutional Neural Networks (SRCNN). Results are described in the Related Publications tab on this page. Wang, Y., Armstrong, R.T., & Mostaghimi, P. (2019). Super Resolution Convolutional Neural Network Models for Enhancing Resolution of Rock Micro-CT Images. arXiv:1904.07470 ( https://arxiv.org/pdf/1904.07470.pdf ) Please cite both this dataset (use its DOI) as well as the related journal publication. Data overview: The dataset is divided into: • 2D and 3D folders • Sandstone and Carbonate Datasets • Combined shuffled Datasets • High Resolution, 2x and 4x downsampled datasets using the Matlab imresize and imresize3 functions with default settings and randomised settings • 2D HR images measure 800x800, cropped samples from the centre of the original cylindrical images, saved as PNG files • 3D HR images measure 80x80x80, cropped from the centre of the original cylindrical images, saved as octave and scipy readable MAT files • 800 images are available for training, 100 for validation, and 100 for testing. • Shuffled subsets contain 1600 images for training, 200 for validation, and 200 for testing The folder structure is as follows: • DRSRD1/ – the parent folder • DRSRD1/DRSRD1_2D/ – folder for all 2D data • DRSRD1/DRSRD1_2D/sandstone2D – folder for 2D sandstone Data • DRSRD1/DRSRD1_2D/carbonate2D– folder for 2D carbonate Data • DRSRD1/DRSRD1_2D/shuffled2D – folder for 2D shuffled Data and similarly for the 3D dataset, found in ./DRSRD1/DRSRD1_3D/ under the exact same structure. Within each carbonate, sandstone, and shuffled folder, the structure is as follows, using the 2D shuffled folder as example: • /shuffled2D_train_HR/ – 0001.png, 0002.png, ..., 0800.png train HR images • /shuffled2D_train_LR_default_X2/ -- 0001x2.png, 0002x2.png, ..., 0800x2.png train LR images, downscale factor x2 • /shuffled2D_train_LR_default_X4/ -- 0001x4.png, 0002x4.png, ..., 0800x4.png train LR images, downscale factor x4 • /shuffled2D_train_LR_unknown_X2/ -- 0001x2.png, 0002x2.png, ..., 0800x2.png train LR images, downscale factor x2 • /shuffled2D_train_LR_unknown_X4/ -- 0001x4.png, 0002x4.png, ..., 0800x4.png train LR images, downscale factor x4 • Validation • /shuffled2D_valid_HR/ -- 0801.png, 0802.png, ..., 0900.png validation HR images • /shuffled2D_valid_LR_default_X2/ -- 0801x2.png, 0802x2.png, ..., 0900x2.png train LR images, downscale factor x2 • /shuffled2D_valid_LR_default_X4/ -- 0801x4.png, 0802x4.png, ..., 0900x4.png train LR images, downscale factor x4 • /shuffled2D_valid_LR_unknown_X2/ -- 0801x2.png, 0802x2.png, ..., 0900x2.png train LR images, downscale factor x2 • /shuffled2D_valid_LR_unknown_X4/ -- 0801x4.png, 0802x4.png, ..., 0900x4.png train LR images, downscale factor x4 • Testing • /shuffled2D_test_HR/ -- 0901.png, 0902.png, ..., 1000.png test HR images • shuffled2D_test_LR_default_X2/ -- 0901x2.png, 0902x2.png, ..., 1000x2.png train LR images, downscale factor x2 • shuffled2D_test_LR_default_X4/ -- 0901x4.png, 0902x4.png, ..., 1000x4.png train LR images, downscale factor x4 • shuffled2D_test_LR_unknown_X2/ -- 0901x2.png, 0902x2.png, ..., 1000x2.png train LR images, downscale factor x2 • shuffled2D_test_LR_unknown_X4/ -- 0901x4.png, 0902x4.png, ..., 1000x4.png train LR images, downscale factor x4