Table 6.

Comparison of different LR configurations on the validation dataset and test dataset.

Configuration	Precision val | test	Recall val | test	mAP50 val | test	mAP50-95 val | test
Default (YOLOv8s+Conv3)	0.899 | 0.931	0.846 | 0.829	0.904 | 0.906	0.674 | 0.694
Freeze–Non–Extra–Conv	0.266 | 0.299	0.308 | 0.352	0.226 | 0.272	0.110 | 0.136
Freeze–Back	0.560 | 0.566	0.363 | 0.461	0.417 | 0.511	0.216 | 0.280
Fast–Extra–Conv	0.883 | 0.906	0.687 | 0.765	0.811 | 0.839	0.545 | 0.603
Fast–Head	0.853 | 0.920	0.750 | 0.778	0.830 | 0.846	0.549 | 0.615
Fast–Head–Neck	0.938 | 0.955	0.830 | 0.837	0.895 | 0.905	0.659 | 0.688

Bold values signify the value of the best performing LR configuration under each metric.