. 2023 Nov 17;13:20117. doi: 10.1038/s41598-023-47398-w

Table 2.

Results on the COCO test-val2017.

Method	Average precision (%)
Method	Size	$A P_{50}$	$A P_{50 : 95}$	$A P_{75}$	$A P_{S}$	$A P_{M}$	$A P_{L}$
RetinaNet¹³	500	53.1	34.4	36.8	14.7	38.5	49.1
SSD¹²	512	48.5	28.8	30.3	10.9	31.8	43.5
CornerNet²⁶	512	57.8	40.5	45.3	20.8	44.8	56.7
CenterNet²⁷	512	62.4	44.9	48.1	25.6	47.4	57.4
YOLOv3¹⁵	608	57.9	33.0	34.3	18.3	35.4	41.9
ASFF³³	608	63.0	42.4	47.4	25.5	45.7	52.3
Efficientdet³²	896	65.0	45.8	49.3	26.6	49.4	59.8
FCOS²⁸	800 $\times$ 1024	64.1	44.7	48.4	27.6	47.5	55.6
DETR-DC5-R101³⁰	800 $\times$ 1333	64.7	44.9	47.7	23.7	49.5	62.3
YOLOv4¹⁷	512	64.9	45.4	48.7	23.0	50.7	62.7
Ours	512	65.4	45.6	48.9	25.4	51.0	62.0
ViDT-tiny³¹	800 $\times$ 1333	64.5	44.8	48.7	25.9	47.6	62.1
ViDT-base³¹	800 $\times$ 1333	69.4	49.2	$53.1$	$30.6$	$52.6$	$66.9$
YOLOv8-S²¹	640	61.8	44.9	–	–	–	–
YOLOv8-L²¹	640	$69.8$	$52.9$	–	–	–	–

The proposed method achieved the best detection results on $A P_{50}$ , which refers to the mean average precision (mAP) for intersection over union (IoU) value thresholds equal to 0.5. $A P_{50 : 95}$ , $A P_{75}$ , $A P_{S}$ , $A P_{M}$ and $A P_{L}$ refer to the average mAP across value of IoU thresholds (i.e., $A P_{50 : 95}$ refers to the average of 10 mAP across IoU thresholds). Bold-italic, italic and boldfaced values represent the best, second-best, and third-best results in each column, respectively.