CAS-YOLO: annotations, trained weights and code for four-stage cucumber growth-stage measurement

This archive contains the annotations, the trained model weights and the source code supporting the paper "Multi-class cucumber growth-stage measurement: a lightweight machine-vision method prioritising the developing stage".Contents 1.data/labels/{train1,train,valid,test}/ - four-stage YOLO-format annotations for 4,274 images and 20,067 instances, including the background-augmented training split (BG-Aux). 2. weights/CAS-YOLO_best.pt - trained CAS-YOLO model (mAP@50 = 90.39%, mAP@50-95 = 70.18% on the validation split). 3.code/ - model configuration, the C2f_Att and ASFF modules, the BG-Aux background extraction script and a training example reproducing the reported setup.The source images are third-party material (publicly released agricultural image datasets and web-collected greenhouse images) and cannot be redistributed here; each annotation file shares the base name of its image, so the images can be matched by file name once obtained from their original providers.