"SPVD-Field: A Multi-source, Multi-task Visual Dataset for Sweet Potato Virus Disease under Real Field Conditions"

"Sweet potato virus disease (SPVD) is a major threat to global sweet potato production, resulting in significant yield losses and posing a serious risk to food security. Vision-based intelligent diagnostic techniques have emerged as a promising solution for large-scale monitoring of SPVD due to their cost-effectiveness and scalability. However, existing publicly available datasets for SPVD are sparse and typically focus on a single task, such as disease classification or lesion segmentation, under controlled imaging conditions. This limits the development, evaluation, and fair comparison of advanced computer vision methods for SPVD analysis.To address this gap, we present SPVD-Field, a comprehensive, multi-source, multi-task visual dataset designed for SPVD research under real field conditions. Unlike existing datasets, which typically focus on a single task, SPVD-Field is organized into two complementary sub-datasets: SPVD-DET for disease detection with bounding box annotations, and SPVD-SEG for fine-grained lesion segmentation with pixel-level masks. These sub-datasets were independently collected using optimized acquisition protocols tailored to their respective tasks, while maintaining a consistent semantic definition of SPVD symptoms, crop growth stages, and field environments.SPVD-Field captures a wide range of real-world variability, including differences in imaging scale, viewpoint, illumination, background complexity, and symptom manifestation, reflecting the challenges of field-based disease diagnosis. Detailed documentation of the data acquisition process, annotation strategies, and quality control procedures are provided. Baseline benchmark results for both detection and segmentation tasks are also included to demonstrate the usability and complexity of the dataset.By offering a structured dataset suite rather than a single-task collection, SPVD-Field aims to support a variety of research directions, including detection, segmentation, multi-task learning, and disease severity analysis, while enabling reproducible and comparable research in SPVD-related plant phenotyping."