StrawDI - The Strawberry Digital Images Data Set

The Strawberry Digital Images (StrawDI) data set contains 8000 images of strawberries, taken from 20 plantations, within an approximate area of 150 hectares, in the province of Huelva, Spain. The plantations were not changed in any way for the experiment and the images were taken from real production conditions during a full picking campaign (from mid-December 2018 to early May 2019). The capture device used was a Samsung Galaxy S7 Edge smartphone attached to an extendable arm. In order to build a data set close to the target application, the images were taken under different conditions of brightness, at a distance of approximately 20 cm from the ridge, at about 35 +-10 cm height and an approximate angle of 25 +-10º. The images have a 4032x3024 resolution and are stored in a JPEG format. StrawDI_Db1 From the total set of images that make up the StrawDI data set, 3100 images have been selected at random. For each of these images a ground truth labelled image has been manually generated containing the individual segmentation of all the strawberries appearing in the image. To reduce the computational demands on the models, the images have been rescaled to 1008 x 756 and stored in PNG format. This set of images constitutes the StrawDI_Db1 database. This database is divided into train (2800 images), validation (100 images) and test (200 images) subsets.草莓数字图像（Strawberry Digital Images，缩写StrawDI）数据集包含8000张草莓图像，采集自西班牙韦尔瓦省（Huelva）的20处种植园，覆盖面积约150公顷。实验期间种植园未做任何改动，图像采集于2018年12月中旬至2019年5月初的完整采摘季，均实拍于真实生产场景。 本次采集采用搭载于可伸缩机械臂上的三星Galaxy S7 Edge智能手机。为使数据集贴合目标应用场景，图像在不同亮度条件下拍摄，拍摄时距种植垄约20厘米，拍摄高度约35±10厘米，拍摄角度约25±10°。图像分辨率为4032×3024，存储格式为JPEG。 StrawDI_Db1 从StrawDI数据集的全部图像中随机选取3100张。针对每张选取的图像，人工生成了对应的真值标注图像，用于对图像中所有草莓个体进行实例分割标注。 为降低模型的计算开销，所有图像均被缩放至1008×756分辨率，并以PNG格式存储。该图像集合构成StrawDI_Db1数据库，该数据库被划分为训练集（2800张）、验证集（100张）与测试集（200张）三个子集。