Attribution Faithfulness on ImageNet (val)
5.21DeletionIntegrated Gradients
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| Integrated GradientsBackbone=ResNet-1012025.04 | 5.21 | 14.82 | 18.22 | |
| LIMA (w/ SLICO)Backbone=ResNet-101, sub-region division algorithm=SLICO2025.04 | 5.67 | 65.38 | 42.03 | |
| LIMA (w/ SLICO)Backbone=MambaVision (L2)2025.04 | 5.7 | 36.8 | — | |
| Previous workBackbone=ResNet-1012025.04 | 6.43 | 50.63 | — | |
| LIMA (w/ SLICO)Backbone=Vision Mamba (base)2025.04 | 8.25 | 69.57 | — | |
| HSIC-AttributionBackbone=ResNet-1012025.04 | 8.66 | 44.48 | 58.05 | |
| Previous workBackbone=MambaVision (L2)2025.04 | 8.72 | 31.21 | — | |
| Integrated GradientsBackbone=Vision Mamba (base)2025.04 | 9.14 | 23.23 | — | |
| Integrated GradientsBackbone=MambaVision (L2)2025.04 | 10.67 | 22.02 | — | |
| RISEBackbone=ResNet-1012025.04 | 11.01 | 49.2 | 38.27 | |
| SaliencyBackbone=ResNet-1012025.04 | 11.48 | 26.13 | 45.56 | |
| Previous workBackbone=Vision Mamba (base)2025.04 | 13.87 | 62.81 | — | |
| RISEBackbone=MambaVision (L2)2025.04 | 14.19 | 29.37 | — | |
| HSIC-AttributionBackbone=MambaVision (L2)2025.04 | 15.41 | 27.53 | — | |
| SaliencyBackbone=MambaVision (L2)2025.04 | 16.81 | 14.94 | — | |
| iGOS++Backbone=ResNet-1012025.04 | 19.34 | 37.79 | 33.77 | |
| Kernel SHAPBackbone=ResNet-1012025.04 | 19.44 | 28.78 | 13.99 | |
| RISEBackbone=Vision Mamba (base)2025.04 | 20.08 | 53.52 | — | |
| Kernel SHAPBackbone=MambaVision (L2)2025.04 | 20.3 | 23.45 | — | |
| HSIC-AttributionBackbone=Vision Mamba (base)2025.04 | 20.67 | 48.72 | — | |
| SaliencyBackbone=Vision Mamba (base)2025.04 | 27.93 | 19.38 | — | |
| Kernel SHAPBackbone=Vision Mamba (base)2025.04 | 29.47 | 36.03 | — |