VeRi776包含超过50,000张776辆车的图像,这些图像由20台摄像机拍摄,在24小时内覆盖1.0平方公里的面积,这使得该数据集可扩展到足以用于车辆Re-Id和其他相关研究。
VehicleID包含白天在中国一个小城市中分布的多个真实监控摄像头捕获的数据。 整个数据集中共有26267辆车(共221763张图像)。 每个图像都带有一个与现实世界中的身份相对应的id标签。 此外,我们手动标记了10319辆车辆(共90196张图像)的车辆型号信息。
VERI-Wild车辆图像由一个包含174个摄像机,拍摄范围覆盖超过200平方公里的市区CCTV系统拍摄。摄像机是24小时连续拍摄30天,其长时间的连续拍摄考虑了车辆真实的各种天气和光照问题。包含了40万张图片,4万种车辆标签。该数据集提供了摄像机ID,时间戳,摄像机之间的跟踪关系。
VRAI使用无人机上的摄像头进行拍摄车辆图像,该数据集包含了137613张图像,包括13022中车辆类别,为了增加车内差异,每辆车至少有两个无人机在不同位置,不同视角和飞行高度(15m至80m)进行拍摄,同时作者手动标记了各种车辆的属性,包括车辆的类别、颜色、天窗、保险杠、备用轮胎和行李架。同时,对于每个车辆图像,还注释标记有差异的部分,用来将特定车辆与其他车辆进行区分。
N-CARS该数据集是基于现实事件的数据集,它由从市区和高速公路环境中的汽车驾驶中获取的约24000个样本组成。用安装在汽车挡风玻璃后面的ATIS摄像机拍摄了80分钟,并将其转换为常规的灰度图像,并标记样本。该数据集有12336个汽车样本和11693非汽车样本组成的两类数据集。其中训练集分为7940个汽车和7482个背景样本,测试集包含4396个汽车样本和4211个背景测试样本。
PKU-VD该数据集包含了两个大型车辆数据集(VD1和VD2),它们分别从两个城市的真实世界不受限制的场景拍摄图像。其中VD1是从高分辨率交通摄像头获得的,VD2中的图像则是从监视视频中获取的。作者对原始数据执行车辆检测,以确保每个图像仅包含一辆车辆。由于隐私保护的限制,所有车牌号码都已被黑色覆盖遮挡。所有车辆图像均从前视图进行拍摄。
The VeRi776 dataset comprises over 50,000 images of 776 vehicles, captured by 20 cameras covering an area of 1.0 square kilometers over a 24-hour period, making this dataset sufficiently scalable for vehicle Re-Id and other related research. The VehicleID dataset includes data captured by multiple real surveillance cameras distributed in a small Chinese city during the day. The entire dataset contains 26,267 vehicles (221,763 images in total). Each image is tagged with an ID corresponding to a real-world identity. Additionally, we have manually labeled the vehicle model information for 10,319 vehicles (90,196 images in total). The VERI-Wild vehicle images were captured by a CCTV system consisting of 174 cameras covering an urban area of over 200 square kilometers. The cameras operated continuously for 30 days, 24 hours a day, considering various weather and lighting conditions for realistic vehicle scenarios. It includes 400,000 images with 40,000 vehicle labels. The dataset provides camera IDs, timestamps, and tracking relationships between cameras. The VRAI dataset uses cameras mounted on drones to capture vehicle images, containing 137,613 images across 13,022 vehicle categories. To increase intra-vehicle variation, each vehicle was captured by at least two drones at different positions, angles, and flight heights (15m to 80m). The authors manually labeled various vehicle attributes, including vehicle type, color, sunroof, bumper, spare tire, and roof rack. Additionally, for each vehicle image, annotated differences are marked to distinguish specific vehicles from others. The N-CARS dataset is based on real events, consisting of approximately 24,000 samples obtained from car driving in urban and highway environments. The samples were captured by an ATIS camera installed behind the car windshield for 80 minutes, converted into regular grayscale images, and labeled. The dataset is divided into two categories: 12,336 car samples and 11,693 non-car samples. The training set includes 7,940 car samples and 7,482 background samples, while the test set contains 4,396 car samples and 4,211 background test samples. The PKU-VD dataset includes two large vehicle datasets (VD1 and VD2), captured from unrestricted real-world scenarios in two cities. VD1 images were obtained from high-resolution traffic cameras, while VD2 images were captured from surveillance videos. The authors performed vehicle detection on the original data to ensure each image contains only one vehicle. Due to privacy protection, all license plate numbers have been obscured with black covers. All vehicle images were captured from the front view.