MultiScene

MultiScene是由德国航空航天中心创建的大型数据集，包含100,000张不受限制的高分辨率航空图像，用于多场景识别研究。数据集通过OpenStreetMap等众包平台进行低成本标注，同时对14,000张图像进行了视觉检查和标签修正，形成了名为MultiScene-Clean的干净标注子集。该数据集不仅提供了地面实况数据，还提供了众包标签，适用于研究从大量噪声标签中学习的问题。MultiScene数据集的应用领域包括环境监测、城市规划、土地覆盖和利用映射以及灾害评估等，旨在解决单一图像中多场景识别的挑战。MultiScene is a large-scale dataset developed by the German Aerospace Center, containing 100,000 unconstrained high-resolution aerial images for multi-scene recognition research. The dataset was annotated at low cost via crowdsourcing platforms such as OpenStreetMap, and 14,000 of the images underwent visual inspection and label correction, forming a clean annotated subset named MultiScene-Clean. This dataset provides both ground truth data and crowdsourced labels, making it suitable for research on learning from large-scale noisy labels. Application areas of the MultiScene dataset include environmental monitoring, urban planning, land cover and utilization mapping, disaster assessment and other fields, aiming to address the challenges of multi-scene recognition in single images.

搜集汇总

背景与挑战

背景概述

MultiScene是一个由德国航空航天中心创建的大型航空图像数据集，包含100,000张高分辨率图像，用于多场景识别研究。它通过众包平台进行低成本标注，并提供了14,000张经过视觉检查和修正的干净子集，既包含地面实况也包含众包标签，适用于研究从噪声标签中学习的问题，应用领域涵盖环境监测、城市规划等。