Table 3.

Detection results on WSD-2022.

Method	Backbone	mAP	AP 50	AP 75	AP m	AP l	#Params (MB)
Faster R-CNN	ResNet50	0.301	0.709	0.215	0.284	0.339	39.4
Mask R-CNN		0.345	0.774	0.237	0.311	0.382	41.9
Faster R-CNN	ResNet101	0.304	0.750	0.208	0.306	0.352	57.6
Mask R-CNN		0.366	0.812	0.246	0.331	0.394	60.1
FCOS	ResNet50	0.368	0.825	0.250	0.355	0.409	30.6
ATSS		0.364	0.803	0.255	0.357	0.402	30.6
SSD	SSDVGG	0.428	0.890	0.362	0.382	0.488	22.7
CenterNet	ResNet18	0.414	0.876	0.318	0.345	0.487	13.8
YOLOv3	DarkNet53	0.437	0.906	0.381	0.387	0.497	58.7
Faster R-CNN	Swin Transformer	0.397	0.881	0.276	0.352	0.450	65.6
Mask R-CNN		0.426	0.914	0.318	0.379	0.473	68.1
FCOS		0.452	0.886	0.402	0.415	0.523	43.8
WheatFormer	MW-Swin Transformer	0.459	0.918	0.384	0.415	0.533	60.1

Faster R-CNN and Mask R-CNN are the representative models of two stages. FCOS, ATSS, SSD, CenterNet, and YOLOv3 are the representative models of one stage.