STEM2Mat-Bench

该数据集是一个专注于二维材料的标准基准数据集，旨在通过STEM图像实现晶体结构的自动化重建。该数据集涵盖了六个化学家族的二维材料，每个结构都以CIF文件形式存储，并附有在真实条件下生成的合成STEM图像。从近10000个候选结构中筛选出了450个定义明确的单层结构。该数据集的任务是利用显微图像实现晶体结构的自动化重建及性质预测。

This is a standard benchmark dataset focused on two-dimensional (2D) materials, dedicated to enabling automated reconstruction of crystal structures via scanning transmission electron microscopy (STEM) images. The dataset covers 2D materials from six chemical families, with each crystal structure stored in Crystallographic Information File (CIF) format and accompanied by synthetic STEM images generated under realistic experimental conditions. A total of 450 well-defined monolayer structures were screened from nearly 10,000 candidate structures. The core tasks of this dataset are automated crystal structure reconstruction and property prediction based on the provided STEM images.