Table 2 Quantitative evaluation on the Vaihingen benchmark.

Method	Backbone	Class F1/IoU %					mF1	mIoU	OA
Method	Backbone	Imp.surf.	Building	Low.veg.	Tree	Car	%	%	%
PSPNet³³	ResNet18	95.19/90.81	94.05/88.77	83.37/71.48	89.60/81.15	82.55/70.28	88.95	80.50	91.58
BiSeNet⁵⁸	ResNet18	95.81/91.96	95.30/91.01	83.96/72.35	89.98/81.78	88.50/79.36	90.71	83.29	92.36
BANet⁵⁹	ResNet18	95.56/91.50	95.24/90.90	83.21/71.25	89.57/81.11	88.60/79.54	90.44	82.86	92.02
A2FPN⁶⁰	ResNet18	95.73/91.81	95.27/90.96	83.48/71.64	89.60/81.16	87.33/77.51	90.28	82.62	92.14
MANet⁶¹	ResNet50	95.77/91.88	95.32/91.06	83.45/71.60	90.02/81.85	88.88/79.99	90.69	83.28	92.25
MAResUNet⁶²	ResNet18	95.72/91.78	95.31/91.04	83.67/71.93	89.78/81.46	87.79/78.23	90.45	82.89	92.19
UNetFormer⁵⁴	ResNet18	95.68/91.72	95.25/90.92	83.85/72.20	89.77/81.43	87.93/78.47	90.50	82.95	92.21
SLCNet⁶³	ResNet50	95.80/91.94	95.47/91.33	84.13/72.61	89.94/81.71	88.93/80.07	90.86	83.53	92.38
GCDNet⁶⁴	ResNet101	95.84/92.01	95.68/91.72	83.65/71.90	89.79/81.47	89.50/81.00	90.89	83.62	92.36
CMTFNet⁶⁴	ResNet50	95.74/91.84	95.93/92.17	84.03/72.45	90.07/81.93	89.40/80.83	91.03	83.84	92.49
SFANet⁶⁵	efficientnet_b3	95.66/91.69	95.70/91.76	83.32/71.41	89.79/81.48	87.64/78.00	90.42	82.87	92.19
MIFNet⁶⁶	ResNeXt	96.03/92.36	95.87/92.07	84.26/72.80	90.10/81.99	89.75/81.40	91.20	84.12	92.66
STUNet⁶⁷	Swin transformer	95.92/92.15	95.63/91.54	84.52/73.12	90.35/82.33	89.21/80.15	91.12	83.85	92.67
DMANet⁶⁸	Swin transformer	96.05/92.41	95.87/91.98	84.89/73.68	90.62/82.77	89.87/80.92	91.46	84.35	92.93
SAM2-ARAFNet (ours)	sam2	96.34/92.94	96.40/93.06	85.70/74.98	91.07/83.61	90.44/82.54	91.99	85.43	93.33

Quick links

Search