Post-hoc Attribution on CLEVR-XAI (val)
29PGIG
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| IGBackbone=ResNet-18, val_acc=91.52%, Mask Type=all_objects, Samples=500, ms/s=131.982026.05 | 29 | 10.9 | 15.1 | 47.9 | 50.3 | |
| VanillaGradBackbone=ResNet-18, val_acc=91.52%, Mask Type=all_objects, Samples=500, ms/s=11.042026.05 | 28.2 | 12.4 | 16.2 | 47.9 | 50.3 | |
| AGOP-LocalBackbone=ResNet-18, val_acc=91.52%, Mask Type=all_objects, Samples=500, ms/s=20.332026.05 | 28.2 | 12.4 | 16.2 | 47.9 | 50.3 | |
| AGOP-WeightedBackbone=ResNet-18, val_acc=91.52%, Mask Type=all_objects, Samples=500, ms/s=18.652026.05 | 20.4 | 12.9 | 16.6 | 48 | 50.4 | |
| RandomBackbone=ResNet-18, val_acc=91.52%, Mask Type=all_objects, Samples=500, ms/s=0.072026.05 | 15.8 | 13 | 15.4 | 47.9 | 50.3 | |
| SmoothGradBackbone=ResNet-18, val_acc=91.52%, Mask Type=all_objects, Samples=500, ms/s=103.502026.05 | 10.8 | 17.9 | 16.9 | 48.2 | 50.5 | |
| GradCAM++Backbone=ResNet-18, val_acc=91.52%, Mask Type=all_objects, Samples=500, ms/s=19.922026.05 | 0.4 | 12.8 | 14.9 | 53.9 | 49.2 | |
| GradCAMBackbone=ResNet-18, val_acc=91.52%, Mask Type=all_objects, Samples=500, ms/s=15.722026.05 | 0.2 | 12.9 | 15.1 | 54.8 | 49.7 | |
| AGOP-GlobalBackbone=ResNet-18, val_acc=91.52%, Mask Type=all_objects, Samples=500, ms/s=0.022026.05 | 0 | 14.9 | 16.5 | 47.9 | 50.4 |