VitiBench Dataset Suite

The VitiBench Dataset Suite is a collection of datasets designed to support computer vision applications in viticulture. It comprises two real-world grapevine leaf image datasets and one synthetic vineyard segmentation dataset. The first leaf dataset contains 472 RGB images representing six Vitis vinifera L. cultivars acquired between May and July 2017 at the University of Trás-os-Montes e Alto Douro experimental vineyards, Portugal, under controlled white-background conditions. The second dataset contains 815 RGB images from twelve grapevine cultivars acquired between July and September 2019 under natural vineyard conditions and varying illumination scenarios. The third dataset was generated using the PROMORE procedural modeling framework, which creates virtual rural environments composed of vineyard rows, trees, buildings, agricultural assets, animals, and low vegetation distributed over Voronoi-based parcels. Simulated aerial image acquisition was performed to automatically generate RGB imagery and corresponding segmentation masks. The dataset suite can be reused in educational activities, benchmarking studies, algorithm validation, and rapid prototyping of classification and semantic segmentation methods for digital ampelography, precision viticulture, and agricultural artificial intelligence.