Looting Detection on Afghanistan looting detection 2023 imagery
0.93AccuracyResNet-50
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| ResNet-50Model Type=CNN2026.02 | 0.93 | 0.915 | 0.94 | 0.926 | 0.97 | |
| ResNet-18Model Type=CNN2026.02 | 0.927 | 0.904 | 0.943 | 0.923 | 0.968 | |
| EfficientNet-B1Model Type=CNN2026.02 | 0.925 | 0.91 | 0.933 | 0.921 | 0.97 | |
| EfficientNet-B0Model Type=CNN2026.02 | 0.923 | 0.913 | 0.923 | 0.918 | 0.966 | |
| ResNet-34Model Type=CNN2026.02 | 0.917 | 0.888 | 0.941 | 0.913 | 0.965 | |
| Handcrafted + XGB + PCAModel Type=Traditional ML, Feature Type=Handcrafted, Classifier=XGBoost, Temporal Aggregation Strategy=PCA2026.02 | 0.718 | 0.703 | 0.678 | 0.69 | 0.786 | |
| SatCLIP-V + RF + MeanModel Type=Traditional ML, Feature Type=SatCLIP-V, Classifier=Random Forest, Temporal Aggregation Strategy=Mean2026.02 | 0.716 | 0.674 | 0.751 | 0.71 | 0.781 | |
| GeoRSCLIP + LR + PCAModel Type=Traditional ML, Feature Type=GeoRSCLIP, Classifier=Logistic Regression, Temporal Aggregation Strategy=PCA2026.02 | 0.69 | 0.662 | 0.674 | 0.668 | 0.751 | |
| Satlas Pretrain + LR + ConcatModel Type=Traditional ML, Feature Type=Satlas Pretrain, Classifier=Logistic Regression, Temporal Aggregation Strategy=Concat2026.02 | 0.623 | 0.591 | 0.61 | 0.599 | 0.676 | |
| SatMAE + GB + ConcatModel Type=Traditional ML, Feature Type=SatMAE, Classifier=Gradient Boosting, Temporal Aggregation Strategy=Concat2026.02 | 0.606 | 0.577 | 0.553 | 0.565 | 0.64 | |
| Prithvi EO 2.0 + LR + PCAModel Type=Traditional ML, Feature Type=Prithvi EO 2.0, Classifier=Logistic Regression, Temporal Aggregation Strategy=PCA2026.02 | 0.597 | 0.563 | 0.57 | 0.566 | 0.635 | |
| DINOv3 + RF + MedianModel Type=Traditional ML, Feature Type=DINOv3, Classifier=Random Forest, Temporal Aggregation Strategy=Median2026.02 | 0.596 | 0.566 | 0.547 | 0.556 | 0.621 |