2019年用于苹果检测和分割的图像数据集

明尼苏达果品数据集由明尼苏达大学某研究团队创建，用于苹果检测。与许多其他专注于单一水果品种或单一生长季节的图像数据集相比，该数据集包含来自多个水果品种在两个生长季节的不同图像。图像数据拍摄在自然光照条件下使用手机摄像头以视频模式完成，然后从记录下的视频序列中提取图像。该数据集分为两组图像，分别用于水果检测和计数任务。检测集由用于训练和测试的两类图片组成，分别包含670张和311张尺寸为1280×720 像素的png格式图像。为训练图像提供使用的像素级水果注释，使用VGG Image Annotator (VIA) 软件制作完成，共代表41,325 个对象实例。计数集由64,595 张jpg 格式图像、2875和3395张png格式图像组成，分别用于训练、验证不同分辨率，并为训练和验证图像提供了从0到6的真实水果计数功能。https://conservancy.umn.edu/handle/11299/206575

The Minnesota Fruit Dataset was developed by a research team at the University of Minnesota for apple detection. Unlike many other image datasets that focus on a single fruit variety or single growing season, this dataset includes diverse images from multiple fruit varieties across two growing seasons. The image data was captured using smartphone cameras in video mode under natural lighting conditions, and images were subsequently extracted from the recorded video sequences. This dataset is split into two image subsets for the fruit detection and counting tasks respectively. The detection subset comprises training and testing image sets, containing 670 and 311 PNG-format images with a resolution of 1280×720 pixels, respectively. Pixel-level fruit annotations are provided for the training images, which were generated using the VGG Image Annotator (VIA) software, totaling 41,325 object instances. The counting subset is composed of 64,595 JPG-format images, 2,875 and 3,395 PNG-format images, which are used for training and validation across different resolutions respectively. Ground-truth fruit counts ranging from 0 to 6 are provided for the training and validation images. https://conservancy.umn.edu/handle/11299/206575