Feature Attribution on ImageNet and COCO
13.2AUC of DeletionExpected Gradients
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| Expected GradientsBackbone=ViT-B/162025.05 | 13.2 | — | — | — | |
| Weighted Integrated GradientsBackbone=ViT-B/162025.05 | 10.6 | — | 19.36 | 0.024 | |
| Expected GradientsBackbone=DenseNet1212025.05 | 8.4 | — | — | — | |
| Weighted Integrated GradientsBackbone=DenseNet1212025.05 | 6.4 | — | 24.03 | 0 | |
| Expected GradientsBackbone=ResNet1012025.05 | 5.7 | — | — | — | |
| Expected GradientsBackbone=ResNet502025.05 | 4.4 | — | — | — | |
| Weighted Integrated GradientsBackbone=ResNet1012025.05 | 3.7 | — | 36.01 | 0 | |
| Expected GradientsBackbone=VGG162025.05 | 3.5 | — | — | — | |
| Weighted Integrated GradientsBackbone=ResNet502025.05 | 2.9 | — | 34.35 | 0.002 | |
| Expected GradientsBackbone=VGG192025.05 | 2.9 | — | — | — | |
| Weighted Integrated GradientsBackbone=VGG162025.05 | 2.7 | — | 23.25 | 0.002 | |
| Weighted Integrated GradientsBackbone=VGG192025.05 | 2.2 | — | 24.63 | 0.018 | |
| Expected GradientsBackbone=MobileNetV22025.05 | 2.1 | — | — | — | |
| Weighted Integrated GradientsBackbone=MobileNetV22025.05 | 1.4 | — | 29.57 | 0.011 | |
| Expected GradientsBackbone=DenseNet1212025.05 | — | 41.8 | — | — | |
| Expected GradientsBackbone=MobileNetV22025.05 | — | 41.4 | — | — | |
| Expected GradientsBackbone=ResNet502025.05 | — | 39 | — | — | |
| Expected GradientsBackbone=ResNet1012025.05 | — | 39.1 | — | — | |
| Expected GradientsBackbone=VGG162025.05 | — | 42.3 | — | — | |
| Expected GradientsBackbone=VGG192025.05 | — | 42.7 | — | — | |
| Expected GradientsBackbone=ViT-B/162025.05 | — | 38.7 | — | — | |
| Weighted Integrated GradientsBackbone=DenseNet1212025.05 | — | 46.2 | 10.59 | 0.001 | |
| Weighted Integrated GradientsBackbone=MobileNetV22025.05 | — | 45.9 | 10.82 | 0.001 | |
| Weighted Integrated GradientsBackbone=ResNet502025.05 | — | 44.8 | 14.85 | 0 | |
| Weighted Integrated GradientsBackbone=ResNet1012025.05 | — | 46.3 | 18.55 | 0 | |
| Weighted Integrated GradientsBackbone=VGG162025.05 | — | 48.8 | 15.33 | 0 | |
| Weighted Integrated GradientsBackbone=VGG192025.05 | — | 45.9 | 7.38 | 0.028 | |
| Weighted Integrated GradientsBackbone=ViT-B/162025.05 | — | 40.9 | 5.69 | 0.046 |