Table 2 Quantitative results on the ISPRS Potsdam dataset. Best results are shown in bold, and second-best results are underlined (%). Imp: impervious surface; Low: low vegetation.
From: Projection Kernel regularization for diffusion-based multimodal remote sensing segmentation
Type | Model | Imp | Building | Low | Trees | Car | OA | F1 | mIoU |
|---|---|---|---|---|---|---|---|---|---|
CNN-based | MANet | 90.61 | 97.58 | 87.34 | 85.45 | 95.61 | 89.82 | 91.05 | 83.91 |
ABCNet | 89.88 | 96.52 | 85.23 | 84.10 | 93.72 | 88.45 | 89.51 | 81.35 | |
PSPNet | 87.53 | 96.04 | 84.61 | 78.99 | 95.14 | 86.70 | 88.01 | 79.07 | |
Transformer-based | FTransUNet | 90.13 | 97.03 | 84.69 | 82.16 | 92.68 | 88.04 | 88.80 | 80.32 |
ASMFNet | 86.46 | 95.78 | 83.03 | 63.63 | 79.78 | 82.36 | 81.15 | 69.19 | |
CMFNet | 91.05 | 97.51 | 86.14 | 79.24 | 95.04 | 88.22 | 88.98 | 80.62 | |
UNetFormer | 91.10 | 97.36 | 88.63 | 85.07 | 95.58 | 89.94 | 91.03 | 83.87 | |
Diffusion-based | SegDiff | 50.48 | 22.83 | 20.71 | 5.51 | 0.21 | 28.03 | 27.39 | 10.71 |
RNDiff | 90.12 | 96.80 | 83.22 | 83.31 | 95.30 | 87.61 | 88.51 | 79.90 | |
PKDiff | 95.12 | 97.72 | 85.76 | 80.17 | 95.33 | 90.54 | 91.02 | 84.03 |
