VessMAP Dataset The VessMAP Dataset comprises 100 manually labeled confocal microscopy images of mouse brain vasculature. Among these images, 20 randomly selected samples were labeled by two annotators. The dataset is organized using the following directory structure: images: 100 blood vessel images (.tiff). Each image has a size of 256x256 pixels with a single grayscale channel of 8-bit color depth. annotator1: labels: 100 labels annotated by annotator 1. Each file is a 256x256 .png binary image, with a value of 255 assigned to pixels annotated as blood vessels and a value of 0 assigned to background pixels. skeletons: skeletons of the labels annotated by annotator 1. Each file is a 256x256 .png binary image, with a value of 255 assigned to pixels belonging to the skeleton and a value of 0 assigned to background pixels. These skeletons were calculated using the Palàgyi-Kuba algorithm [1]. measures.json: a .json file with four metrics that can be used to map the dataset to a feature space. The four metrics included in the file are the contrast, blood vessel density, estimation of Gaussian noise level, and skeleton heterogeneity. A detailed explanation of the metrics can be found in [2]. These metrics are useful for testing possible biases when training supervised segmentation algorithms. annotator2: This directory contains annotations for 20 images that were randomly selected from the 100 images of the whole dataset. The organization is the same as for annotator 1. 1. K. Palágyi and A. Kuba, "A 3D 6-subiteration thinning algorithm for extracting medial lines," Pattern Recognit. Lett. 19, 613–627 (1998). 2. da Silva, MV., Santos, NdC., Lacoste, B., & Comin, CH. A new sampling methodology for creating rich, heterogeneous, subsets of samples for training image segmentation algorithms. arXiv preprint arXiv:2301.04517, (2023).