Semantic Segmentation on Wind Blade Segmentation (test)
98.25AccuracyMARG+RegionMix
Evaluation results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| MARG+RegionMixRegion classifier architecture=EfficientNet-B4, Perfect classifier assumption=false2026.01 | 98.25 | 98.43 | 96.49 | 97.67 | 95.33 | — | — | |
| SAM2Perfect classifier assumption=true2026.01 | 97.86 | 97.46 | 94.82 | 95.84 | 94.82 | — | — | |
| CSDA(ln)2026.01 | 97.4 | 95.72 | 97.38 | 96.17 | 93.72 | 94.25 | 93.18 | |
| BU-Net2026.01 | 97.39 | 99.42 | 93.35 | 95.73 | 93.8 | 94.7 | 92.9 | |
| BU-NetRegion classifier architecture=EfficientNet-B4, Perfect classifier assumption=false2026.01 | 97.39 | 99.42 | 93.35 | 95.73 | 93.8 | — | — | |
| Mask2Former-FreqFusionRegion classifier architecture=EfficientNet-B4, Perfect classifier assumption=false2026.01 | 97.35 | 98.02 | 96.12 | 96.67 | 94.65 | — | — | |
| CSDA(∆)dcs=42026.01 | 97.34 | 95.97 | 97.3 | 96.2 | 93.94 | 94.51 | 93.38 | |
| CSDA2026.01 | 96.98 | 96.78 | 96.65 | 96.13 | 93.86 | 94.21 | 93.51 | |
| Mask2Former2026.01 | 96.68 | 95.63 | 93.89 | 94.76 | 93.72 | 94.49 | 92.93 | |
| Mask2FormerRegion classifier architecture=EfficientNet-B4, Perfect classifier assumption=false2026.01 | 96.68 | 95.63 | 93.89 | 94.76 | 93.72 | — | — | |
| EfficientFormer2026.01 | 96.42 | 95.47 | 93.63 | 94.55 | 93.51 | 94.02 | 92.99 | |
| EfficientFormerRegion classifier architecture=EfficientNet-B4, Perfect classifier assumption=false2026.01 | 96.42 | 95.47 | 93.63 | 94.55 | 93.81 | — | — | |
| DiffSeg2026.01 | 96.37 | 82.08 | 89.74 | 85.73 | 86.4 | 91.66 | 81.13 | |
| DiffSegPerfect classifier assumption=true2026.01 | 96.37 | 89.74 | 85.73 | 86.4 | 72.14 | — | — | |
| U-NetFormer2026.01 | 96.2 | 97.31 | 93.51 | 94.42 | 91.75 | 92.53 | 90.96 | |
| U-NetFormerRegion classifier architecture=EfficientNet-B4, Perfect classifier assumption=false2026.01 | 96.2 | 97.31 | 93.51 | 94.42 | 91.75 | — | — | |
| MobileViT2026.01 | 96.14 | 95.44 | 93.33 | 94.38 | 93.47 | 94.06 | 92.88 | |
| MobileViTRegion classifier architecture=EfficientNet-B4, Perfect classifier assumption=false2026.01 | 96.14 | 95.44 | 93.33 | 94.38 | 93.47 | — | — | |
| BiRefNetRegion classifier architecture=EfficientNet-B4, Perfect classifier assumption=false2026.01 | 95.65 | 98.37 | 92.52 | 94.57 | 92.38 | — | — | |
| SAM2026.01 | 94.36 | 97.29 | 91.22 | 92.6 | 91.66 | 92.31 | 91.01 | |
| SAMPerfect classifier assumption=true2026.01 | 94.36 | 97.29 | 91.22 | 92.6 | 91.66 | — | — | |
| ResNeSt2026.01 | 94.23 | 96.84 | 91.47 | 92.77 | 89.63 | 90.4 | 88.86 | |
| ResNeStRegion classifier architecture=EfficientNet-B4, Perfect classifier assumption=false2026.01 | 94.23 | 96.84 | 91.47 | 92.77 | 89.63 | — | — | |
| DeepLabv3+2026.01 | 94.14 | 96.36 | 87.38 | 89.03 | 87.47 | 90.31 | 84.62 | |
| DeepLabv3+Region classifier architecture=EfficientNet-B4, Perfect classifier assumption=false2026.01 | 94.14 | 96.36 | 87.38 | 89.03 | 87.47 | — | — | |
| SW2026.01 | 93.48 | 93.57 | 91.71 | 91.37 | 87.44 | 88.64 | 86.23 | |
| SWRegion classifier architecture=EfficientNet-B4, Perfect classifier assumption=false2026.01 | 93.48 | 93.57 | 91.71 | 91.37 | 87.44 | — | — | |
| CLIPSeg2026.01 | 82.7 | 77.02 | 75.52 | 74.29 | 75.09 | 80.16 | 70.02 | |
| CLIPSegRegion classifier architecture=EfficientNet-B4, Perfect classifier assumption=false2026.01 | 82.7 | 77.02 | 75.52 | 74.29 | 75.09 | — | — | |
| PDDA(ln)Loss Type=Log-normal2026.01 | 0.9797 | 0.9466 | 0.9871 | 0.9628 | 0.9434 | 0.9528 | 0.934 | |
| PDDA(∆)Loss Type=Delta2026.01 | 0.9762 | 0.9429 | 0.9779 | 0.9553 | 0.9329 | 0.9436 | 0.9221 | |
| BU-Net2026.01 | 0.9739 | 0.9942 | 0.9335 | 0.9573 | 0.938 | 0.947 | 0.929 | |
| Mask2Former2026.01 | 0.9668 | 0.9563 | 0.9389 | 0.9476 | 0.9372 | 0.9451 | 0.9293 | |
| EfficientFormer2026.01 | 0.9642 | 0.9547 | 0.9363 | 0.9455 | 0.9351 | 0.9402 | 0.9299 | |
| DiffSeg2026.01 | 0.9637 | 0.8208 | 0.8974 | 0.8573 | 0.864 | 0.9166 | 0.8113 | |
| MobileViT2026.01 | 0.9614 | 0.9544 | 0.9333 | 0.9438 | 0.9347 | 0.9406 | 0.9288 | |
| SAMMode=Zero-shot2026.01 | 0.9436 | 0.9729 | 0.9122 | 0.926 | 0.9166 | 0.9231 | 0.9101 | |
| ResNeSt2026.01 | 0.9423 | 0.9684 | 0.9147 | 0.9277 | 0.8963 | 0.904 | 0.8886 | |
| DeepLabv3+2026.01 | 0.9414 | 0.9636 | 0.8738 | 0.8903 | 0.8747 | 0.9031 | 0.8462 | |
| SW2026.01 | 0.9348 | 0.9357 | 0.9171 | 0.9137 | 0.8744 | 0.8864 | 0.8623 | |
| U-NetRole=Baseline backbone2026.01 | 0.8624 | 0.9551 | 0.6893 | 0.7795 | 0.7594 | 0.8031 | 0.7157 | |
| CLIPSeg2026.01 | 0.827 | 0.7702 | 0.7552 | 0.7429 | 0.7509 | 0.8016 | 0.7002 |