CERTH Grape Dataset

The CERTH grape dataset aims to advance computer vision and machine learning research in the field of viticulture by providing valuable annotated data for developing and refining algorithms for accurate grape segmentation, yield prediction, and, most importantly, maturity estimation. The dataset consists of 2502 high-resolution images captured from a vineyard cultivating the 'Crimson Seedless' table grape variety during the 2022–2023 development and harvesting period and exhibits different view angles, camera focus conditions, and illumination variations. The images are captured using an iPhone 11 Pro smartphone and are scaled to a resolution of 2160 × 3840 pixels. The annotations include grape bunches with detailed object masks and are categorized into three distinct classes (i.e., immature, semi-mature, and mature) based on the degree of grape maturity, as identified by the color of the grapes in the bunch. As a result, grapes in the immature class were early in their development phase, grapes in the mature class were close to the harvesting season, and grapes in the semi-mature class were in the intermediate period when changes in the color of the grapes from yellow to red had initiated. The CERTH grape dataset consists of 9832 labeled grape bunches, extracted from all 2502 images. The grape images are split into training, validation, and test sets, consisting of 2000, 251, and 251 images, respectively. The training set comprises 7959 annotated grape bunches, while the validation and test sets comprise 914 and 959 annotated grape bunches, respectively. The CERTH grape datasets can be evaluated on two scenarios. The multi-instance / multi-class (mimc) scenario evaluates the performance of algorithms in multi-instance classification using the original test set, while the single-instance / one-class (sioc) scenario extracts 100 images from the original test set that contain a single grape bunch per image to evaluate the performance of algorithms in single-instance classification. The sioc subset consists of 36, 43, and 21 images that depict grape bunches from the immature, semi-mature, and mature classes, respectively. To use this dataset, please also cite: Blekos, A.; Chatzis, K.; Kotaidou, M.; Chatzis, T.; Solachidis, V.; Konstantinidis, D.; Dimitropoulos, K. A Grape Dataset for Instance Segmentation and Maturity Estimation. Agronomy 2023, 13, 1995. https://doi.org/10.3390/agronomy13081995.

CERTH葡萄数据集旨在推动葡萄栽培领域的计算机视觉与机器学习研究，通过提供高价值的标注数据，助力开发与优化用于精准葡萄分割、产量预测，以及尤为重要的成熟度估测的算法。该数据集包含2502张高分辨率图像，采集自2022-2023年生长与收获季的"Crimson Seedless"（克瑞森无核）鲜食葡萄种植园，涵盖不同拍摄视角、相机对焦状态与光照变化条件。所有图像均通过iPhone 11 Pro智能手机拍摄，并统一调整至2160×3840像素的分辨率。

标注内容包含带有精细目标掩码的葡萄果串，并依据果串中葡萄的颜色所判定的成熟度等级，划分为三个类别：未成熟、半成熟与成熟。其中，未成熟类别的葡萄处于生长早期阶段，成熟类别的葡萄已临近收获季，而半成熟类别的葡萄则处于葡萄颜色由黄色向红色转变的过渡中期。该数据集共包含9832个标注葡萄果串，均提取自全部2502张图像。

该葡萄图像集被划分为训练集、验证集与测试集，分别包含2000、251与251张图像。训练集涵盖7959个标注葡萄果串，验证集与测试集则分别包含914与959个标注果串。

CERTH葡萄数据集可在两种评估场景下开展评估。多实例多分类（multi-instance / multi-class, MIMC）场景采用原始测试集，用于评估算法在多实例分类任务中的性能；单实例单分类（single-instance / one-class, SIOC）场景则从原始测试集中提取100张每张仅包含单个葡萄果串的图像，以评估算法在单实例分类任务中的性能。SIOC子集分别包含36、43与21张对应未成熟、半成熟、成熟类别的葡萄果串图像。

若使用该数据集，请引用以下文献：Blekos, A.; Chatzis, K.; Kotaidou, M.; Chatzis, T.; Solachidis, V.; Konstantinidis, D.; Dimitropoulos, K. 面向实例分割与成熟度估测的葡萄数据集. Agronomy, 2023, 13, 1995. https://doi.org/10.3390/agronomy13081995.

搜集汇总

背景与挑战

背景概述

CERTH Grape Dataset是一个专为葡萄栽培研究设计的数据集，包含2502张高分辨率图像和9832个标注葡萄串，分为三个成熟度类别。数据集支持实例分割和成熟度估计任务，适用于计算机视觉和机器学习算法的开发和评估。