Attribution Map Faithfulness on FunnyBirds (test)
97.6AccuracyChefer-LRP
Evaluation results
| Method | Links | ||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Chefer-LRPBackbone=ViT-B-16/224, Pre-training=ImageNet-21k2025.07 | 97.6 | 91.1 | 91.2 | 89.4 | 89.7 | — | 99.8 | 73.9 | — | 95.8 | — | 86.6 | |
| GenericBackbone=ViT-B-16/224, Pre-training=ImageNet-21k2025.07 | 97.6 | 91 | 90.8 | 89.6 | 89.6 | — | 99.8 | 74.2 | — | 98.5 | — | 87.6 | |
| IIABackbone=ViT-B-16/224, Pre-training=ImageNet-21k2025.07 | 97.6 | 89.2 | 87.6 | 88 | 90.7 | — | 99.8 | 76.4 | — | 98.6 | — | 84.1 | |
| ViT-CXBackbone=ViT-B-16/224, Pre-training=ImageNet-21k2025.07 | 97.6 | 56.9 | 36.2 | 41.6 | 83.8 | — | 99.8 | 78.3 | — | 57.7 | — | 66.8 | |
| IBABackbone=ViT-B-16/224, Pre-training=ImageNet-21k2025.07 | 97.6 | 96 | 97.8 | 94.4 | 91.8 | — | 99.8 | 76.9 | — | 71.7 | — | 80.8 | |
| BeyondBackbone=ViT-B-16/224, Pre-training=ImageNet-21k2025.07 | 97.6 | 87.8 | 84.8 | 84.8 | 84.1 | — | 99.8 | 75.8 | — | 92.9 | — | 84.5 | |
| CoIBABackbone=ViT-B-16/224, Pre-training=ImageNet-21k, departure layer s=4, arrival layer e=122025.07 | 97.6 | 93.5 | 94.2 | 91.4 | 91.3 | — | 99.8 | 79 | — | 98.2 | — | 89.8 |