TomatoPGT: A 3D point cloud dataset of tomato plants for segmentation and plant-trait extraction

TomatoPGT is a three-dimensional (3D) point cloud dataset of greenhouse-grown tomato plants designed to support research in semantic segmentation, graph-based plant representation, and plant trait extraction. The dataset comprises three tomato cultivars: Celebrity Hybrid, Beefsteak, and Big Boy Hybrid, each captured at multiple developmental stages using a rotational multi-view imaging system with controlled illumination. For each plant and stage, 60–70 overlapping RGB images were acquired and reconstructed into metrically scaled 3D point clouds using structure-from-motion and multi-view stereo techniques.

The dataset provides RGB images, dense colored point clouds, semantic and instance annotations, semantic graph representations encoding plant topology, and tabular phenotypic traits derived from the graph structures. Graph representations describe the structural relationships between plant organs, including stem segments, junctions, stalks, and compound leaves, enabling organ-resolved geometric and architectural analysis.