VessMAP
收藏资源简介:
VessMAP是由圣卡洛斯联邦大学计算机科学系等机构创建的异质性血血管分割数据集,包含100张从18279张未标注图像中精心选取的图像。该数据集通过独特的采样方法,确保了图像的多样性,适用于评估分割算法在数据分布变化下的性能。VessMAP的创建旨在解决医学图像分割中模型偏见的问题,特别是在处理不常见或异常样本时的挑战。数据集的应用领域包括开发新的分割算法,以及在少样本和主动学习中的应用。
VessMAP is a heterogeneous blood vessel segmentation dataset developed by the Department of Computer Science of the Federal University of São Carlos and other institutions. It consists of 100 carefully selected images sourced from 18,279 unannotated images in total. By adopting a unique sampling approach, the dataset guarantees the diversity of image samples, which makes it applicable for evaluating the performance of segmentation algorithms under data distribution shifts. The development of VessMAP targets resolving the problem of model bias in medical image segmentation, particularly the challenges posed by uncommon or anomalous samples. Its application scenarios include the development of novel segmentation algorithms, as well as its utilization in few-shot learning and active learning.




