Label-Free Coreset Selection with Foundation Models for Efficient Annotation in Computational Pathology

作者：Tuo Yin, Jennifer Dhont · 单位：LABEL-FREE CORESET SELECTION WITH FOUNDA-, Department of Medical Physics, Institut Jules Bordet, Hˆopital Universitaire de Bruxelles (HUB), Belgium, Radiophysics and MRI Physics Laboratory, Universit´e Libre de Bruxelles (ULB), Belgium, Laboratory of Image Synthesis and Analysis (LISA), Universit´e Libre de Bruxelles (ULB), Belgium, tice. In this study, we present GCcore, a novel label-free coreset selection method, positions depending on their hyperparameter settings. Code is publicly available., peting label-free selection methods across 10, et al., 2023), motivating a fundamental question: given an unlabeled histopathology dataset of N, access to prior label information remains challenging, and, to date, no systematic and reproducible, access to labels and only subsequently adapted to the label-free regime. Existing approaches can, histopathology datasets for two main reasons. First, existing label-free coreset selection methods, single class label · 会议/期刊：arXiv preprint · 方向：cs.CV · 发布日期：2026-10-06