Infrared Small Target Detection on SIRST-UAVB
27.91mIoUMSHNet
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| MSHNetScale loss variant=L1 (Diff-based), Gaussian-shaped spatial attention (GP)=true, Learnable rotated pinwheel mask=true2026.04 | 27.91 | 73.9 | 8.85 | |
| MSHNetScale loss variant=L4 (Var-denominator), Gaussian-shaped spatial attention (GP)=true, Learnable rotated pinwheel mask=true2026.04 | 27.71 | 69.37 | 8.37 | |
| SCTransNet2026.04 | 27.33 | 74.34 | 8.87 | |
| MSHNetScale loss variant=L2 (Var-based), Gaussian-shaped spatial attention (GP)=true, Learnable rotated pinwheel mask=true2026.04 | 27.09 | 68.97 | 6.82 | |
| MSHNetScale loss variant=L3 (Mobius), Gaussian-shaped spatial attention (GP)=true, Learnable rotated pinwheel mask=true2026.04 | 26.89 | 69.84 | 8.48 | |
| UIU-Net2026.04 | 26.84 | 66.57 | 7.38 | |
| MSHNetScale loss variant=L3 (Mobius)2026.04 | 26.72 | 73.92 | 9.09 | |
| MSHNetScale loss variant=L3 (Mobius), Gaussian-shaped spatial attention (GP)=true2026.04 | 26.65 | 70.56 | 9.09 | |
| MSHNetScale loss variant=L2 (Var-based), Gaussian-shaped spatial attention (GP)=true2026.04 | 26.22 | 72.03 | 8.21 | |
| MSHNetScale loss variant=L4 (Var-denominator)2026.04 | 25.82 | 72 | 10.54 | |
| MTU-Net2026.04 | 25.74 | 65.49 | 8.82 | |
| MSHNetScale loss variant=L2 (Var-based)2026.04 | 24 | 66.42 | 8.98 | |
| DNANet2026.04 | 23.96 | 63.72 | 15.11 | |
| MSHNetScale loss variant=L1 (Diff-based)2026.04 | 23.8 | 63.48 | 7.21 | |
| MSHNetScale loss variant=L1 (Diff-based), Gaussian-shaped spatial attention (GP)=true2026.04 | 23.58 | 64.5 | 8.87 | |
| MSHNetScale loss variant=L1 (Diff-based), Location regularizer=omitted, Gaussian-shaped spatial attention (GP)=true2026.04 | 22.37 | 61.92 | 9.65 | |
| MSHNetScale loss variant=L3 (Mobius), Location regularizer=omitted, Gaussian-shaped spatial attention (GP)=true2026.04 | 22 | 67.5 | 9.59 | |
| MSHNetScale loss variant=L2 (Var-based), Location regularizer=omitted, Gaussian-shaped spatial attention (GP)=true2026.04 | 20.7 | 60.72 | 10.76 | |
| ACM2026.04 | 18.46 | 69.24 | 11.08 | |
| MSHNetScale loss variant=L4 (Var-denominator), Gaussian-shaped spatial attention (GP)=true2026.04 | 16.84 | 66.57 | 9.38 | |
| MSHNetScale loss variant=L4 (Var-denominator), Location regularizer=omitted, Gaussian-shaped spatial attention (GP)=true2026.04 | 15.95 | 52.65 | 9.21 |