Table 4.

Model performance comparison of YOLOv8 models using SGD and Adam optimizers.

Model	Size	Optimizer	Best Epoch	mAP${}^{\mathit{val}}$ 50	mAP${}^{\mathit{val}}$ 50-95	Speed GPU RTX 3080Ti (ms)
YOLOv8s	640	SGD	56	0.611	0.389	4.4
YOLOv8s	640	Adam	57	0.604	0.383	4.3
YOLOv8s	1024	SGD	36	0.623	0.395	5.4
YOLOv8s	1024	Adam	47	0.625	0.399	4.9
YOLOv8m	640	SGD	52	0.621	0.396	4.9
YOLOv8m	640	Adam	62	0.621	0.403	5.5
YOLOv8m	1024	SGD	35	0.624	0.402	9.9
YOLOv8m	1024	Adam	70	0.626	0.401	10.0

For training with the SGD optimizer, the initial learning rate is 1$\times {10}^{-2}$; for training with the Adam optimizer, the initial learning rate is 1$\times {10}^{-3}$.