Table 3 Quantitative evaluation on the potsdam benchmark.
Method | Backbone | Class F1/IoU % | mF1 | mIoU | OA | ||||
|---|---|---|---|---|---|---|---|---|---|
Imp.surf. | Building | Low.veg. | Tree | Car | % | % | % | ||
PSPNet33 | ResNet18 | 92.57/86.17 | 94.29/89.20 | 86.07/75.55 | 86.76/76.62 | 94.45/89.49 | 90.83 | 83.41 | 89.61 |
BiSeNet58 | ResNet18 | 93.77/88.27 | 96.07/92.43 | 87.00/76.99 | 88.39/79.20 | 96.04/92.39 | 92.25 | 85.86 | 91.07 |
BANet59 | ResNet18 | 93.32/87.48 | 95.95/92.21 | 86.65/76.45 | 88.61/79.54 | 95.78/91.90 | 92.06 | 85.52 | 90.73 |
A2FPN60 | ResNet18 | 93.33/87.49 | 95.58/91.54 | 86.76/76.62 | 88.22/78.92 | 95.76/91.86 | 91.93 | 85.28 | 90.73 |
MANet61 | ResNet50 | 93.88/88.47 | 96.42/93.08 | 87.16/77.25 | 88.77/79.81 | 96.03/92.36 | 92.45 | 86.19 | 91.22 |
MAResUNet61 | ResNet18 | 93.44/87.69 | 96.19/92.65 | 86.88/76.80 | 88.28/79.02 | 95.73/91.81 | 92.10 | 85.59 | 90.82 |
UNetFormer54 | ResNet18 | 90.86/83.24 | 93.11/87.10 | 82.99/70.93 | 82.08/69.60 | 93.23/87.32 | 88.45 | 79.64 | 87.03 |
SLCNet63 | ResNet50 | 93.04/86.98 | 95.84/92.01 | 86.80/76.68 | 88.81/79.87 | 95.61/91.58 | 92.02 | 85.42 | 90.66 |
GCDNet64 | ResNet101 | 93.97/88.62 | 96.36/92.98 | 87.13/77.19 | 88.62/79.56 | 95.57/91.52 | 92.33 | 85.98 | 91.24 |
CMTFNet69 | ResNet50 | 93.80/88.32 | 96.54/93.32 | 87.81/78.28 | 88.82/79.89 | 96.14/92.57 | 92.63 | 86.48 | 91.38 |
SFANet65 | efficientnet_b3 | 93.75/88.24 | 96.46/93.17 | 86.86/76.77 | 88.50/79.38 | 95.87/92.07 | 92.29 | 85.93 | 90.98 |
MIFNet66 | ResNeXt | 94.18/89.00 | 97.05/94.28 | 87.31/77.49 | 89.17/80.46 | 96.53/93.31 | 92.85 | 86.90 | 91.55 |
STUNet67 | Swin Transformer | 94.05/88.73 | 96.23/92.81 | 87.42/77.65 | 89.05/80.12 | 96.18/92.61 | 92.59 | 86.81 | 91.42 |
DMANet68 | Swin Transformer | 94.28/89.15 | 96.47/93.25 | 87.76/78.12 | 89.38/80.64 | 96.35/92.89 | 92.85 | 86.81 | 91.68 |
SAM2-ARAFNet (ours) | Sam2 | 94.69/89.92 | 97.10/94.37 | 88.42/79.24 | 89.53/81.05 | 96.18/92.64 | 93.18 | 87.44 | 92.13 |
