Explanation Quality Evaluation on COVID-Qu-Ex (test)
37.7RRA (FeatPerm)DenseNet-169
Evaluation results
| Method | Links | ||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|
| DenseNet-169Backbone=DenseNet-1692026.05 | 37.7 | 28.5 | 22.3 | 22.6 | 29.6 | 61.4 | 12.3 | 50 | 49.9 | 49.5 | |
| ResNet-34Backbone=ResNet-342026.05 | 34.7 | 21.9 | 23.3 | 23.5 | 29 | 57.6 | 11 | 49.7 | 49.8 | 49.9 | |
| ResNet-18Backbone=ResNet-182026.05 | 34.5 | 25.8 | 22.8 | 22.8 | 28.3 | 57 | 12.2 | 50.1 | 50 | 49.9 | |
| DenseNet-201Backbone=DenseNet-2012026.05 | 34.4 | 32.3 | 20.6 | 21 | 26.5 | 60.3 | 13.4 | 50 | 50.1 | 49.7 | |
| DenseNet-121Backbone=DenseNet-1212026.05 | 33.6 | 22.1 | 20.9 | 21.1 | 28.3 | 60.2 | 10.7 | 49.7 | 49.7 | 49.7 | |
| ViT-B/16Backbone=ViT-B/16, Resolution=224, Pre-training=IN-21k2026.05 | 30.2 | 19.5 | 21.3 | 21.1 | 29.3 | 51 | 9.7 | 49.5 | 49.5 | 49.7 | |
| ResNet-101Backbone=ResNet-1012026.05 | 30.1 | 25.3 | 22.6 | 22.6 | 28.6 | 54.2 | 11.4 | 50.2 | 50.3 | 49.7 | |
| ResNet-50Backbone=ResNet-50, Pre-training=IN-1k2026.05 | 29.2 | 26.6 | 19.8 | 20.1 | 24.7 | 54.1 | 12 | 49.9 | 49.9 | 50 | |
| ViT-L/16Backbone=ViT-L/16, Resolution=224, Pre-training=IN-21k2026.05 | 29 | 20.3 | 21.4 | 21.4 | 27.8 | 51.3 | 10.1 | 49.9 | 49.8 | 49.9 | |
| ResNet-50Backbone=ResNet-50, Pre-training=scratch2026.05 | 27.9 | 22.9 | 21.6 | 21.4 | 28.8 | 53.4 | 11 | 49.9 | 49.8 | 49.7 | |
| DenseNet-121Backbone=DenseNet-121, Pre-training=CheXpert, Convergence=did not converge within the training budget2026.05 | 25.8 | 34.9 | 22.5 | 22.8 | 29.4 | 47.8 | 14.9 | 50 | 50 | 49.7 |