Feature Attribution on ImageNet 1000 samples
57.44Insertion ScoreFAMPE
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| FAMPEArchitecture=MaxViT-T, Alpha selection strategy=per-sample best α2025.09 | 57.44 | 10.15 | |
| FAMPEArchitecture=MaxViT-T, Alpha selection strategy=fixed α = 0.12025.09 | 52.1 | 14.8 | |
| FAMPEArchitecture=Inception-v3, Alpha selection strategy=per-sample best α2025.09 | 48.02 | 5.95 | |
| AGIArchitecture=MaxViT-T2025.09 | 47.92 | 21.84 | |
| AttEXploreArchitecture=MaxViT-T2025.09 | 46.5 | 17.53 | |
| FAMPEArchitecture=Inception-v3, Alpha selection strategy=fixed α = 0.12025.09 | 43.88 | 10.15 | |
| AttEXploreArchitecture=Inception-v32025.09 | 42.09 | 9.99 | |
| AGIArchitecture=Inception-v32025.09 | 41.43 | 6.53 | |
| FAMPEArchitecture=ResNet-50, Alpha selection strategy=per-sample best α2025.09 | 39.02 | 9.53 | |
| AGIArchitecture=ResNet-502025.09 | 35.72 | 5.43 | |
| AttEXploreArchitecture=ResNet-502025.09 | 34.91 | 12.56 | |
| FAMPEArchitecture=VGG-16, Alpha selection strategy=per-sample best α2025.09 | 32.89 | 8.29 | |
| IGArchitecture=MaxViT-T2025.09 | 30.44 | 23.46 | |
| FAMPEArchitecture=ResNet-50, Alpha selection strategy=fixed α = 0.12025.09 | 29.78 | 14.4 | |
| AGIArchitecture=VGG-162025.09 | 29.35 | 4.31 | |
| AttEXploreArchitecture=VGG-162025.09 | 28.81 | 9.62 | |
| GIGArchitecture=MaxViT-T2025.09 | 28.37 | 20.91 | |
| DeepLiftArchitecture=MaxViT-T2025.09 | 25.63 | 24.13 | |
| FAMPEArchitecture=VGG-16, Alpha selection strategy=fixed α = 0.12025.09 | 22.49 | 13.46 | |
| GIGArchitecture=Inception-v32025.09 | 18.21 | 3.13 | |
| IGArchitecture=Inception-v32025.09 | 16.77 | 3.69 | |
| DeepLiftArchitecture=Inception-v32025.09 | 15.43 | 5.05 | |
| GIGArchitecture=ResNet-502025.09 | 9.07 | 1.92 | |
| IGArchitecture=ResNet-502025.09 | 8.58 | 2.65 | |
| DeepLiftArchitecture=ResNet-502025.09 | 8.46 | 2.96 | |
| GIGArchitecture=VGG-162025.09 | 6.41 | 2.53 | |
| DeepLiftArchitecture=VGG-162025.09 | 6.12 | 2.95 | |
| IGArchitecture=VGG-162025.09 | 5.73 | 2.94 |