COOS-7

COOS-7（Cells Out Of Sample 7-Class）是由多伦多大学创建的一个公共数据集，包含132,209张小鼠细胞图像。该数据集旨在通过四个测试数据集，评估图像分类器在不同程度的协变量偏移下的泛化能力。数据集内容包括一个包含41,456张图像的训练数据集和四个测试数据集，每个数据集代表了与训练数据集不同的变化因素。COOS-7数据集的创建过程涉及从更大的显微镜实验数据集中精选高质量图像，并通过不同的显微镜和实验条件进行复现。该数据集主要应用于生物医学领域，特别是在评估和改进图像分类器在实际应用中的泛化能力方面。COOS-7 (Cells Out Of Sample 7-Class) is a public dataset developed by the University of Toronto, which contains 132,209 mouse cell images. The core goal of this dataset is to evaluate the generalization ability of image classifiers under varying degrees of covariate shift through four test datasets. Specifically, the dataset includes a training subset with 41,456 images and four test subsets, each of which embodies a unique covariate shift factor distinct from the training dataset. The construction of COOS-7 involves screening high-quality images from a larger-scale microscopy experimental dataset and replicating the samples under different microscopes and experimental conditions. This dataset is mainly utilized in the biomedical domain, particularly for assessing and enhancing the generalization performance of image classifiers in real-world applications.

搜集汇总

背景与挑战

背景概述

COOS-7是一个包含132,209张小鼠细胞图像的公共数据集，由多伦多大学创建，用于评估图像分类器在不同协变量偏移下的泛化能力。数据集包括一个训练集和四个测试集，每个测试集代表不同的变化因素，如显微镜和实验条件，主要应用于生物医学领域，以改进图像分类器的实际应用性能。