Flood risk classification on BCUR dataset (test)
94.7AccuracySkjold-DiT
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| Skjold-DiTBackbone=24-layer transformer, Diffusion Steps=50, Sampler=DDIM2026.02 | 94.7 | 93 | 6.7 | |
| UrbanDiTType=Diffusion-transformer mobility foundation model2026.02 | 91.8 | 90 | 10.4 | |
| GraphSAGEType=Graph Neural Network2026.02 | 89.4 | 87 | 12.8 | |
| CNN-ResNet50Backbone=ResNet-50, Type=CNN-based imagery model2026.02 | 88.2 | 86 | 14.3 | |
| Random ForestType=Classical ML baseline2026.02 | 84.7 | 82 | 18.9 | |
| HAND-DEMType=Physics-inspired baseline2026.02 | 76.3 | 71 | 31.2 |