无人机拍摄图像数据集（VisDrone2019）

无人机，或一般的无人机，配备了摄像头，已经被快速部署到广泛的应用中，包括农业，航空摄影，快速交付和监控。因此，自动理解从这些平台上收集的视觉数据变得非常重要，这使得计算机视觉与无人机的关系越来越密切。我们很高兴地提出了一个大规模的基准，为各种重要的计算机视觉任务提供了精心注释的地面真相，命名为VisDrone，以使视觉与无人机相遇。 VisDrone2021数据集是由中国天津大学机器学习和数据挖掘实验室的AISKYEYE团队收集。该基准数据集由400个视频片段组成，包括265,228个帧和10,209个静态图像，由各种无人机安装的相机拍摄，涵盖了广泛的方面，包括位置（取自中国相隔数千公里的14个不同城市）、环境（城市和乡村）、物体（行人、车辆、自行车等）和密度（稀疏和拥挤的场景）。请注意，该数据集是使用不同的无人机平台（即不同型号的无人机），在不同的场景，以及不同的天气和照明条件下收集的。这些帧是由260多万个边界框或经常感兴趣的目标点手动注释的，如行人、汽车、自行车和三轮车。为了更好地利用数据，还提供了一些重要的属性，包括场景可见度、物体类别和遮挡。 Zhu P, Wen L, Du D, et al. Detection and Tracking Meet Drones Challenge[J]. IEEE Transactions on Pattern Analysis & Machine Intelligence, 2021 (01): 1-1. Bibtex source | Abstract | PDF

Unmanned aerial vehicles (UAVs), or drones in general, equipped with cameras, have been rapidly deployed across a wide range of applications, including agriculture, aerial photography, rapid delivery, and surveillance. As a result, automatically understanding the visual data collected from these platforms has become critically important, which has increasingly tied computer vision closely to drones. We are pleased to present a large-scale benchmark named VisDrone, which provides meticulously annotated ground truths for various important computer vision tasks, to bridge the gap between computer vision and drones. The VisDrone2021 dataset was collected by the AISKYEYE Team from the Machine Learning and Data Mining Laboratory of Tianjin University, China. This benchmark dataset consists of 400 video clips, including 265,228 frames and 10,209 static images, captured by various on-board drone cameras, covering a wide range of aspects, including locations (collected from 14 different cities across China thousands of kilometers apart), environments (urban and rural areas), objects (pedestrians, vehicles, bicycles, etc.), and densities (sparse and crowded scenes). It should be noted that this dataset was collected using different drone platforms (i.e., various drone models), in diverse scenarios, and under varying weather and lighting conditions. These frames were manually annotated with over 2.6 million bounding boxes and frequently targeted points of interest, such as pedestrians, cars, bicycles, and tricycles. To facilitate better utilization of the dataset, several critical attributes are also provided, including scene visibility, object categories, and occlusion. Zhu P, Wen L, Du D, et al. Detection and Tracking Meet Drones Challenge[J]. IEEE Transactions on Pattern Analysis & Machine Intelligence, 2021 (01): 1-1. Bibtex source | Abstract | PDF

搜集汇总

背景与挑战

背景概述

VisDrone2019数据集是一个大规模的无人机拍摄图像和视频数据集，包含来自中国14个不同城市的265,228帧和10,209张静态图像，涵盖多种场景和条件，并提供了详细的标注信息，适用于多种计算机视觉任务。