BU-BIL (Boston University Biomedical Image Library)

BU-BIL 是一个图像库，包括六个数据集，代表三种成像模式和六种对象类型。指示数据集的提供者选择能够捕捉他们研究中出现的各种环境条件和成像噪声的图像。然后要求这些专家从那些反映这些物体可以表现出的形状和外观的自然多样性的图像中选择物体。裁剪包含已识别对象的图像子区域以创建图像库。结果是一个包含来自 235 个图像的 305 个对象的库。作者通过视觉检查验证图像库包括各种对象外观、背景和区分对象与背景的属性。论文：如何收集生物医学图像的分割？评估专家、众包非专家和算法绩效的基准

BU-BIL is an image library comprising six datasets, representing three imaging modalities and six object categories. The dataset providers were instructed to select images that capture various environmental conditions and imaging noise encountered in their research. Experts were then asked to select objects from images that reflect the natural diversity of shapes and appearances that these objects can exhibit. Sub-regions of the images containing the identified objects were cropped to create the image library. The resulting library contains 305 objects from 235 images. The authors verified via visual inspection that the library includes diverse object appearances, backgrounds, and attributes that distinguish objects from their backgrounds. Paper: How to Collect Segmentation Datasets for Biomedical Images? A Benchmark for Evaluating Expert, Crowdsourced Non-Expert, and Algorithmic Performance

搜集汇总

背景与挑战

背景概述

BU-BIL是一个生物医学图像数据集，包含来自235张图像的305个对象，涵盖六种对象类型和三种成像模式。该数据集由专家选取以捕捉自然多样性，并于2014年由波士顿大学发布。