SWM dataset: A large-scale annotated benchmark for Sclerotinia sclerotiorum object detection in white mold management

Computer vision enables automated, accurate plant disease scouting, especially for diseases with patchy field distributions and lifecycles that begin in the soil and evolve across soil and plants. This is the case for the soilborne fungal pathogen Sclerotinia sclerotiorum, the causal agent of white mold in over 400 hosts, such as common bean, soybean, cotton, sunflower, canola, and tobacco, threatening sustainable agriculture on all continents. There are image datasets focused on plant disease detection, including the Plant Disease Database (PDDB) and PlantDoc, which cover a large range of plant diseases. None of them considers real-field images of S. sclerotiorum and white mold for detection. The novel dataset fills this gap by providing images of common bean crops annotated by plant pathologists, covering three life cycle stages: dormant sclerotia, germinated apothecia, and white mold symptoms. The image set is structured into two datasets: the SWM Dataset and the R-SWM Dataset. The SWM Dataset consists of a preprocessed and class-balanced version, whereas the R-SWM Dataset preserves the original raw-field images. Both datasets provide comprehensive annotations in YOLO and Pascal VOC formats, facilitating their adoption across different object detection models. The datasets are split into training, validation, and test sets, enabling the development of deep learning-based detection approaches to improve detector performance, address related challenges, and advance research in computer vision and plant pathology.