CaryaData: an image dataset of Chinese hickory in natural orchards

CaryaData is an open image dataset for fruit maturity assessment of Chinese hickory (Carya cathayensis Sarg.) collected in natural orchard environments in Lin’an District, Hangzhou, Zhejiang Province, China. The dataset comprises 1,661 canopy images at a resolution of 3024 × 3024 pixels, covering key developmental stages from fruit enlargement to harvest under diverse illumination conditions and background clutter. Within these images, 3,211 visually discernible fruit instances are annotated with axis-aligned bounding boxes and assigned to three maturity levels (maturity1–maturity4) or a uknown class for visually uncertain cases, based on pericarp colour, surface blemishes and cracking status. Data construction followed a multi-stage quality control workflow including manual image screening, standardised maturity interpretation guidelines and double-annotator spot checks to ensure consistency and reliability. The dataset is organised into four main components: (1) Raw – 1,661 original orchard images at 3024 × 3024 pixels; (2) Scaled – images uniformly resized to 640 × 640 pixels with correspondingly updated annotations; (3) Metadata – image-level and instance-level CSV files describing fruit counts, maturity composition, illumination and geometric properties; and (4) CaryaData_YOLO – a ready-to-use version derived from the Scaled subset with train/val/test splits (6:2:2) compatible with mainstream object detection frameworks. Annotations are provided in YOLO (TXT), PASCAL VOC (XML) and COCO (JSON) formats, all generated from a single manually curated source. CaryaData supports a range of computer-vision tasks, including multi-class fruit detection, maturity classification from cropped instances, analysis of class imbalance and few-shot learning, and studies of illumination variation, occlusion and background interference in natural orchards. Baseline experiments with YOLOv8n demonstrate that the dataset is suitable for deep learning-based maturity detection and provides a reusable benchmark for research on maturity evaluation, yield estimation and intelligent harvesting of Chinese hickory.