IPSM-Bench
收藏资源简介:
IPSM-Bench是由北京科技大学等机构联合创建的大规模锌合金微观结构图像数据集,旨在解决中间相分割领域标注数据稀缺的挑战。该数据集包含1,054张高分辨率(512×512)扫描电子显微镜和光学显微镜图像,覆盖20种锌基合金体系,共计22,179个经专家手动标注的中间相实例,数据源自实验制备和学术文献。其构建过程采用激光粉末床熔融技术制备样品,并结合AI辅助预标注与人工校验流程,确保标注精确性和一致性。该数据集主要应用于材料科学和生物医学工程领域,为锌基可吸收生物材料的微观结构分析、中间相自动分割模型训练与性能评估提供关键基础,助力优化合金力学与功能特性。
IPSM-Bench is a large-scale zinc alloy microstructure image dataset jointly developed by the University of Science and Technology Beijing and other institutions, aiming to address the challenge of scarce labeled data in the field of intermetallic phase segmentation. This dataset contains 1,054 high-resolution (512×512) images obtained via scanning electron microscopy and optical microscopy, covering 20 zinc-based alloy systems, and includes a total of 22,179 intermetallic phase instances manually annotated by experts. The data is sourced from both experimentally prepared samples and academic literature. In the construction process, laser powder bed fusion technology was used to fabricate the samples, and a workflow combining AI-assisted pre-annotation and manual verification was adopted to ensure the accuracy and consistency of the annotations. This dataset is primarily applied in the fields of materials science and biomedical engineering, providing a critical foundation for microstructure analysis of zinc-based absorbable biomedical materials, training and performance evaluation of automated intermetallic phase segmentation models, and assisting in optimizing the mechanical and functional properties of alloys.
数据集名称
IPSM-Bench(Intermediate Phase Segmentation Benchmark in Microstructure Images of Zinc-based Absorbable Biomaterials)
数据集简介
IPSM-Bench是一个用于锌合金微观结构中中间相分割的高质量基准数据集,发表于IJCAI 2026。
数据采集
- 图像数量:共1,054张图像,其中800张为自行构建,254张来源于学术文献。
- 图像类型:涵盖光学显微镜(OM)和扫描电子显微镜(SEM)两种模态。
- 采集流程:包括样品制备、金相制备和金相图像采集三个阶段。
数据标注
采用AI辅助标注策略,包含三个关键阶段:
- 基于AI模型的预标注
- 人工验证与修正
- 一致性检查
数据统计
- 合金体系:包含20种锌基合金体系。
- 图像分布:
- ZnCu:370张(占35.10%)
- ZnCuMg:269张(占25.52%)
- 两者合计占60.63%。
- 标注实例:总计22,179个。
- ZnCu:12,355个(占55.71%)
- ZnCuMg:4,705个(占21.21%)
- 两者合计占76.92%。
- 模态分布(训练/测试):
- 训练集:554张SEM图像,246张OM图像。
- 测试集:122张SEM图像,132张OM图像。
数据集下载
- 使用限制:仅限学术研究使用,禁止再分发、商业利用或超出签署协议范围的使用。
- 获取方式:研究人员需填写并签署Release Agreement,然后将签署后的协议发送至指定联系邮箱。
- 联系邮箱:xujinglin@ustb.edu.cn, yagengli@ustb.edu.cn, m202520912@xs.ustb.edu.cn
- 引用要求:使用数据必须引用相关论文。
引用格式
bibtex @inproceedings{IPSM-Bench_ijcai2026, title = {IPSM-Bench: A New Intermediate Phase Segmentation Benchmark in Microstructure Images of Zinc-based Absorbable Biomaterials}, author = {Jinglin Xu, Shangyan Zhao, Jiabo Wang, Xinghong Mu, Yulong Lei, Jiacheng Zhang, Hongbo Sun and Yageng Li}, booktitle = {Proceedings of the 35th International Joint Conference on Artificial Intelligence}, year = {2026} }




