TomatoWUR: an annotated dataset of 3D tomato plants to quantitatively evaluate segmentation, skeletonisation, and plant trait extraction algorithms for 3D plant phenotyping

In this research, we created a dataset of 44 annotated 3D point clouds of tomato plants. Images were captured using fifteen cameras surrounding a single plant. Those images were used to create a point cloud using the shape-from silhouette methodology (Golbach et al., 2016). The resulting dataset was annotated and separated into three parts. 1) Annotated point clouds and corresponding RGB images with semantic and instance labels , 2) annotated skeletons to analyse plant architecture, and 3) manual reference measurements of internode length, internode diameter, leaf angle, and phyllotactic angle to evaluate phenotyping algorithms from point cloud to plant traits.

本研究构建了包含44份带标注番茄植株三维点云的数据集。实验采用环绕单株番茄植株的15台相机采集图像，基于采集所得的图像通过剪影形状重建法（shape-from silhouette）生成点云（Golbach等，2016）。所获数据集经标注后分为三个部分：1）带有语义标签与实例标签的标注点云及对应RGB图像；2）用于分析植株构型的标注骨架；3）用于评估从点云提取植株性状的表型分析算法的手动参考测量数据，涵盖节间长度、节间直径、叶夹角与叶序角。