Table 1. Comparison of detection results using SSD and YOLOv3 series models.

Model	Smin	Smax	mAP	Blast	PRO	MYE	MET	bNEU	sNEU	LYM	MON	EO	BA	rLYM
SSD300×300	0.2	0.9	0.931	0.970	0.870	0.862	0.881	0.982	0.970	0.884	0.914	1.0	0.993	0.913
	0.1	0.9	0.831	0.882	0.654	0.745	0.663	0.853	0.928	0.849	0.806	0.985	0.973	0.800
SSD512×512	0.2	0.9	0.687	0.651	0.453	0.682	0.499	0.777	0.741	0.409	0.756	0.907	0.891	0.790
YOLOv3_320×320 -			0.925	0.961	0.826	0.860	0.841	0.955	0.968	0.937	0.912	0.999	0.993	0.920
YOLOv3_416×416		-	0.921	0.958	0.824	0.862	0.826	0.947	0.968	0.947	0.894	0.999	0.987	0.915
YOLOv3_608×608		-	0.919	0.962	0.814	0.864	0.829	0.961	0.974	0.930	0.904	0.999	0.987	0.881
YOLOv3-tiny_416×416 -			0.772	0.941	0.469	0.701	0.372	0.758	0.915	0.871	0.674	0.993	0.982	0.815

SSD300×300 has an input size of 300×300, SSD512×512 increases it to 512×512, and YOLOv3 is the 3rd version of YOLO architecture, with the different backbone net (YOLOv3-tiny) and different input size: 320×320, 416×416, and 608×608.