Multi-dataset of grapevine leaf disc images for Downy Mildew severity estimation

This dataset contains three collections of grapevine leaf disc images associated with expert assessments of downy mildew (Plasmopara viticola) severity. The datasets were collected and curated to support the development and evaluation of automated methods for disease severity estimation in digital plant pathology. The collection comprises 774 cropped leaf disc images from three datasets differing in acquisition conditions, image resolution, visual appearance, and severity distribution. Dataset 1 contains 505 images, Dataset 2 contains 153 images, and Dataset 3 contains 116 images. The datasets include images covering a broad range of disease severity, although the distributions are skewed toward low-severity samples. Disease severity was assessed independently by five experts on a 0–100 scale following established guidelines for the assessment of Plasmopara viticola. The image data consist of grapevine leaf discs showing different levels of downy mildew infection. Each image is associated with severity assessments provided by multiple human experts, enabling the study of inter-expert variability and the use of aggregated expert assessments as reference values for automated estimation. Additionally, the raw DINOv2 embeddings used for statistical analysis are included. The collection is intended for research on grapevine downy mildew, disease severity estimation, digital plant pathology, automated phenotyping, computer vision, representation learning, and explainable artificial intelligence.