Fig. 3 | Scientific Reports

Fig. 3

From: Successes and limitations of pretrained YOLO detectors applied to unseen time-lapse images for automated pollinator monitoring

Fig. 3

Grid search results for the optimal NMS confidence and NMS-IoU hyperparameters for YOLO detectors (localisation task, independent frames), with a focus on the maximum F1 score (panels a and b) and area under the precision-recall curve (AUC, panels c and d). The YOLOv5-small model demonstrates superior performance (highest F1 and AUC), achieving optimal detection at an NMS confidence estimate of 0.2019 (panel b) and a NMS-IoU of 0.3 (panels a and c), marked with grey dotted vertical lines. The presented F1-confidence curve (panel b) and the precision-recall curve (panel d) correspond to the optimal NMS-IoU for each model. The evaluation was performed using an eval-IoU of 0.5.

Back to article page