PoreBoostGAN

The MRCCM dataset, accessible via the Digital Rocks Portal(https://www.digitalrocksportal.org/projects/362), comprises multi-resolution digital rock scans of two distinct carbonate samples, each imaged at two different resolutions. This open-access dataset is particularly valuable for advancing research in digital rock analysis, including the study of partial volume effects, segmentation techniques, and the benchmarking of deep learning models for super-resolution and segmentation tasks. We specifically selected two digital rocks from the MRCCM dataset for validation purposes.

To ensure compatibility with the Biogenic dataset—which features scans of more complex and long-term geological structures—the MRCCM dataset has been converted to .tif format. The Biogenic dataset includes three in-situ resolution scans, enabling direct validation of ultra-high-resolution models against ground truth data. In contrast, the MRCCM dataset, with only two available resolutions, has undergone a downsampling strategy to generate the necessary data for comprehensive training and evaluation.

The SR folder within the dataset is particularly noteworthy. It contains data specifically prepared for testing the extrapolation capabilities of super-resolution models, making it an essential resource for researchers focused on developing and refining these models. The root directory of the dataset includes 3D reconstruction training sets for digital rock cores, while the SR folder provides datasets tailored for evaluating the performance of super-resolution techniques in extrapolating higher-resolution details.

This structure ensures that researchers have access to both the foundational data necessary for training and specialized datasets designed for rigorous testing of super-resolution methods.

MRCCM数据集可通过数字岩石门户（Digital Rocks Portal，https://www.digitalrocksportal.org/projects/362）获取，包含两块不同碳酸盐岩样品的多分辨率数字岩心扫描数据，每块样品均以两种不同分辨率完成成像。该开放获取数据集对推动数字岩石分析领域的研究具有重要价值，涵盖部分体积效应研究、分割技术研发，以及面向超分辨率与分割任务的深度学习模型基准测试等方向。我们特意从MRCCM数据集中选取了两份数据用于验证工作。

为兼容生物成因数据集（Biogenic dataset）——该数据集包含更复杂的长期地质结构扫描数据——MRCCM数据集已被转换为.tif格式。生物成因数据集包含三份原位分辨率扫描样本，可实现超高分辨率模型与真值数据的直接验证对比。与之相对，仅具备两种分辨率的MRCCM数据集通过下采样策略生成了用于全面训练与评估的必要数据。

数据集中的SR文件夹尤为值得关注，其中包含专为测试超分辨率模型外推能力而制备的专用数据，是专注于开发与优化此类模型的研究人员的核心资源。数据集根目录包含数字岩心的三维重建训练集，而SR文件夹则提供了专为评估超分辨率技术外推更高分辨率细节性能打造的专用数据集。

该数据集的结构设计确保研究人员既能获取用于模型训练的基础数据，也可获得用于严格测试超分辨率方法的专用测试数据集。

搜集汇总

背景与挑战

背景概述

PoreBoostGAN数据集是一个多分辨率数字岩石扫描数据集，包含两个碳酸盐样本的两种分辨率扫描数据，主要用于数字岩石分析和深度学习模型研究。数据集特别提供了SR文件夹，用于测试超分辨率模型的外推能力，为相关研究提供了重要的数据支持。