Sugarcane single eye-bud detection: annotated image datasets and trained detector weights (cultivars COJN9509 and Co86032)

Annotated image datasets and trained object-detection weights. The in-domain dataset (cultivar CoJN-9509; 1,502 RGB images; train/valid/test = 1053/224/225) provides hand-drawn bounding boxes for three classes -EYE_BUD, NODE and INTERNODE ; in Ultralytics YOLO format. An external cross-variety test set (cultivar Co-86032; 414 test images) is included for zero-retraining generalisation. Trained weights include the full-data YOLO26n specialist detector (eye-bud AP@0.5 ~ 0.92; runs in ~15 ms on a CPU-only Raspberry Pi 5) and two other nano architectures. Supplementary CSVs contain the label-efficiency and annotation-scheme results reported in the paper. See README.md for full structure, class definitions and reproduction notes.