Heritage Science

Table 3 Performance comparison among different segmentation methods

From: Ancient mural segmentation based on multiscale feature fusion and dual attention enhancement

Model	Backbone	MIoU/%	MPA/%	FPS
FCN	VGG16	77.36
SegNet	VGG16	82.23	92.84	17.26
PSPNet	ResNet50	85.07	93.8	22.17
DeepLabV3 +	Xception	85.75	94.37	41.54
MC-DM	MobileNetV2	84.42	93.71	56.52
DANet	ResNet50	85.2	94.53	67.31
SETR	ViT-L	85.14	94.2	15.65
DAFPN	Swin-T	86.82	95.25	26.8
MFAM (Ours)	CA_MobileViT	88.19	95.66	45.43

The bold number indicates the highest value among the evaluation indicators

Back to article page