Fig. 3.

Training and validation results of YOLOv8 models using our hardwood stomatal image dataset (a), and the model performance (b) and (c). Train/box_loss, train/cls_loss, train/dfl_loss indicate the bounding boxes loss, class loss, and distribution focal loss, respectively, during the training process; Val/box_loss, val/cls_loss, val/dfl_loss represent the bounding boxes loss, class loss, and distribution focal loss, respectively, during the validation process; metrics/mAP50(B), metrics/recall(B) represent the model evaluation metrics, the mAP50 represents mean average precision at intersection over union (IOU) = 0.50, B is used to distinguish the metrics of segmentation (i.e., metrics/precision(B) for detection and metrics/precision(M) for segmentation).