HRTEM-specific noise calibration dataset
收藏资源简介:
该数据集由北京理工大学团队开发,专为高分辨率透射电子显微镜(HRTEM)成像中的成核过程观测设计。其核心包含模拟的无序原子排列结构和经过HRTEM专用噪声校准的真实图像噪声,能够有效支持深度学习模型在原子定位任务中的去噪性能。数据生成过程采用创新的噪声校准技术,通过空间域和频域统计特征引导,精确复现了HRTEM快速成像中的复杂噪声分布。该数据集主要应用于先进固体材料研究领域,旨在解决非晶-晶体转变过程中原子级动态观测的噪声干扰问题,为理解石墨烯、金属催化剂等材料的形成机制提供关键数据支持。
This dataset was developed by a research team from Beijing Institute of Technology, and is specifically tailored for the observation of nucleation processes in high-resolution transmission electron microscopy (HRTEM) imaging. The core of this dataset includes simulated disordered atomic arrangement structures and real image noise that has been specifically calibrated for HRTEM, which effectively supports the optimization and performance evaluation of deep learning models for denoising tasks in atomic localization. The dataset generation process utilizes an innovative noise calibration technique, which is guided by statistical features in both spatial and frequency domains to accurately reproduce the complex noise distribution present in fast HRTEM imaging. This dataset is primarily applied in the field of advanced solid materials research, with the goal of addressing noise interference issues in atomic-scale dynamic observations during amorphous-crystal transition processes, thereby providing critical data support for understanding the formation mechanisms of materials such as graphene and metal catalysts.
SCGN 数据集概述
数据集基本信息
- 数据集名称:SCGN
- 关联论文:Statistical Characteristic-Guided Denoising for Rapid High-Resolution Transmission Electron Microscopy Imaging
- 论文链接:https://arxiv.org/abs/2603.18834
- 会议/年份:IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2026年
数据集内容与用途
- 核心用途:用于训练和测试SCGN(Statistical Characteristic-Guided Denoising)模型,该模型旨在实现快速高分辨率透射电子显微镜成像的去噪。
- 测试数据集:包含在
tem_test_data4.zip压缩包中。 - 训练数据集:名为
tem_data4.zip,需通过指定链接下载(提取码:f198)。
数据集使用流程
- 解压测试数据集
tem_test_data4.zip。 - 从指定网盘下载并解压训练数据集
tem_data4.zip。 - 运行
train_convLast_std_tem_data4.py脚本进行模型训练与测试。网络架构定义在convLast_std.py文件中。 - 可视化结果将保存在
convLast_std_tem_data4_result文件夹中。程序将输出PSNR、SSIM和IOU指标结果。 - 预训练模型权重文件为
convLast_std_tem_data4_100.pth,测试时会自动加载。如需重新训练,可删除此文件。
相关资源
- 网络架构代码:
convLast_std.py - 训练与测试脚本:
train_convLast_std_tem_data4.py - 预训练模型权重:
convLast_std_tem_data4_100.pth
引用信息
如需在研究中引用,请使用提供的BibTeX格式引用关联论文。




