BugNIST

BugNIST是一个用于对象检测领域转移的大型体积数据集，由丹麦技术大学创建。该数据集包含9154个微CT体积的12种不同类型的昆虫和388个体积的紧密包装的昆虫混合物。数据集的特点是源域和目标域中的对象外观相同，这在其他领域转移基准数据集中是不常见的。在训练过程中，使用单独扫描的昆虫体积进行标注，而测试则使用带有中心点标注和昆虫类型标签的混合物。BugNIST旨在解决体积3D图像中的对象检测和分类问题，特别是在对象周围上下文变化而对象外观保持不变的情况下。该数据集的应用领域包括医学成像和其他基于微CT的领域，如材料微观结构分析和特殊生物医学应用。

BugNIST is a large volumetric dataset for domain adaptation in object detection, created by the Technical University of Denmark. This dataset contains 9,154 micro-CT volumes of 12 distinct insect species, plus 388 volumes of tightly packed insect mixtures. A key characteristic of BugNIST is that the object appearances in both source and target domains are identical, which is uncommon among other domain adaptation benchmark datasets. During training, individually scanned insect volumes are used for annotation, while the test set employs mixtures with center-point annotations and insect category labels. BugNIST is designed to address object detection and classification tasks in volumetric 3D images, particularly in scenarios where the surrounding context of objects varies while their appearances remain unchanged. Application scenarios of this dataset include medical imaging and other micro-CT-based fields, such as material microstructure analysis and specialized biomedical applications.

搜集汇总

背景与挑战

背景概述

BugNIST是一个大型3D CT扫描数据集，包含9542个昆虫体积扫描，涵盖12种昆虫类型，分为9154个单独扫描和388个混合物扫描，混合物标注中心点用于对象检测。该数据集专门设计用于研究域转移问题，即在单独昆虫上训练的模型应用于混合物检测时的挑战，由于没有使用限制，适合作为深度学习在体积图像分析中的基准数据集。