Explainability on ImageNet (val)
72.67InsertionRISE
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| RISEBackbone=ResNet502018.06 | 72.67 | 10.76 | — | — | |
| LIMEBackbone=ResNet502018.06 | 69.4 | 12.17 | — | — | |
| Grad-CAMBackbone=ResNet502018.06 | 67.66 | 12.32 | — | — | |
| RISEBackbone=VGG162018.06 | 66.63 | 9.8 | — | — | |
| Sliding windowBackbone=ResNet502018.06 | 66.18 | 14.21 | — | — | |
| LIMEBackbone=VGG162018.06 | 61.67 | 10.14 | — | — | |
| Grad-CAMBackbone=VGG162018.06 | 61.49 | 10.87 | — | — | |
| Sliding windowBackbone=VGG162018.06 | 59.17 | 11.58 | — | — | |
| RISEBackbone=ResNet50, Access=Black-box2022.06 | 55.4 | 11.4 | — | 8.427 | |
| RISEBackbone=ResNet50V2, Method Type=Black box2021.11 | 54.6 | — | — | — | |
| VarGradBackbone=ResNet50, Access=White-box2022.06 | 52.7 | 13.4 | — | 0.097 | |
| GradCAMBackbone=ResNet50V2, Method Type=White box2021.11 | 52.4 | — | — | — | |
| GradCAM++Backbone=ResNet50, Access=White-box2022.06 | 49.7 | 12.4 | — | 0.127 | |
| RISEBackbone=VGG16, Access=Black-box2022.06 | 48.5 | 10.6 | — | — | |
| RISEBackbone=VGG16, Method Type=Black box2021.11 | 48.4 | — | — | — | |
| H_s^2 acc.Backbone=ResNet50, Access=Black-box, p_samples=1536, Grid size=7x72022.06 | 48.1 | 10.5 | — | 1.668 | |
| Kernel ShapBackbone=ResNet50, Access=Black-box2022.06 | 48 | 18.5 | — | 4.097 | |
| LIMEBackbone=ResNet50, Access=Black-box2022.06 | 47.2 | 18.6 | — | 6.48 | |
| H_s^2 eff.Backbone=ResNet50, Access=Black-box, p_samples=764, Grid size=7x72022.06 | 47 | 10.6 | — | 0.956 | |
| SmoothGradBackbone=ResNet50V2, Method Type=White box2021.11 | 44.5 | — | — | — | |
| RISEBackbone=MobileNetV2, Method Type=Black box2021.11 | 44.3 | — | — | — | |
| RISEBackbone=MobileNetV2, Access=Black-box2022.06 | 44.3 | 11.5 | — | — | |
| RISEBackbone=EfficientNet, Method Type=Black box2021.11 | 43.9 | — | — | — | |
| RISEBackbone=EfficientNet, Access=Black-box2022.06 | 43.9 | 11.3 | — | — | |
| GradCAMBackbone=VGG16, Method Type=White box2021.11 | 43.8 | — | — | — | |
| GradCAMBackbone=MobileNetV2, Method Type=White box2021.11 | 41.9 | — | — | — | |
| GradCAM++Backbone=VGG16, Access=White-box2022.06 | 41.3 | 12.5 | — | — | |
| VarGradBackbone=MobileNetV2, Access=White-box2022.06 | 39.9 | 9.7 | — | — | |
| H_s^2 acc.Backbone=VGG16, Access=Black-box, p_samples=1536, Grid size=7x72022.06 | 39.5 | 9.9 | — | — | |
| GradCAMBackbone=EfficientNet, Method Type=White box2021.11 | 39.3 | — | — | — | |
| Kernel ShapBackbone=VGG16, Access=Black-box2022.06 | 39.3 | 16.5 | — | — | |
| H_s^2 acc.Backbone=MobileNetV2, Access=Black-box, p_samples=1536, Grid size=7x72022.06 | 39.2 | 9.3 | — | — | |
| GradCAM++Backbone=MobileNetV2, Access=White-box2022.06 | 38.7 | 10.6 | — | — | |
| H_s^2 eff.Backbone=VGG16, Access=Black-box, p_samples=764, Grid size=7x72022.06 | 38.7 | 10 | — | — | |
| Integ.-Grad.Backbone=ResNet50, Access=White-box2022.06 | 38.6 | 13.8 | — | 1.024 | |
| LIMEBackbone=MobileNetV2, Access=Black-box2022.06 | 38.4 | 14.8 | — | — | |
| Kernel ShapBackbone=MobileNetV2, Access=Black-box2022.06 | 38.3 | 14.9 | — | — | |
| H_s^2 eff.Backbone=MobileNetV2, Access=Black-box, p_samples=764, Grid size=7x72022.06 | 38.1 | 9.4 | — | — | |
| SmoothGradBackbone=ResNet50, Access=White-box2022.06 | 37.9 | 12.7 | — | 0.063 | |
| Guided-Backprop.Backbone=ResNet50V2, Method Type=White box2021.11 | 37.7 | — | — | — | |
| SmoothGradBackbone=VGG16, Method Type=White box2021.11 | 37.4 | — | — | — | |
| Sobol (STi)Backbone=ResNet50V2, Method Type=Black box2021.11 | 37 | — | — | — | |
| SobolBackbone=ResNet50, Access=Black-box2022.06 | 37 | 12.1 | — | 5.254 | |
| Kernel ShapBackbone=EfficientNet, Access=Black-box2022.06 | 36.7 | 16.4 | — | — | |
| H_s^2 acc.Backbone=EfficientNet, Access=Black-box, p_samples=1536, Grid size=7x72022.06 | 36.6 | 9.4 | — | — | |
| SaliencyBackbone=ResNet50V2, Method Type=White box2021.11 | 36.3 | — | — | — | |
| Grad.-InputBackbone=ResNet50, Access=White-box2022.06 | 36.3 | 15.3 | — | 0.023 | |
| Guided-Backprop.Backbone=MobileNetV2, Method Type=White box2021.11 | 36.1 | — | — | — | |
| SaliencyBackbone=ResNet50, Access=White-box2022.06 | 35.7 | 15.8 | — | 0.36 | |
| H_s^2 eff.Backbone=EfficientNet, Access=Black-box, p_samples=764, Grid size=7x72022.06 | 35.7 | 9.5 | — | — | |
| Sobol (STi)Backbone=MobileNetV2, Method Type=Black box2021.11 | 33.1 | — | — | — | |
| SobolBackbone=MobileNetV2, Access=Black-box2022.06 | 33.1 | 10.7 | — | — | |
| GradCAM++Backbone=EfficientNet, Access=White-box2022.06 | 31.6 | 11.2 | — | — | |
| Sobol (STi)Backbone=VGG16, Method Type=Black box2021.11 | 31.3 | — | — | — | |
| SobolBackbone=VGG16, Access=Black-box2022.06 | 31.3 | 10.9 | — | — | |
| Sobol (STi)Backbone=EfficientNet, Method Type=Black box2021.11 | 30.9 | — | — | — | |
| SobolBackbone=EfficientNet, Access=Black-box2022.06 | 30.9 | 10.4 | — | — | |
| DeconvNetBackbone=ResNet50V2, Method Type=White box2021.11 | 30.7 | — | — | — | |
| SmoothGradBackbone=MobileNetV2, Method Type=White box2021.11 | 30.7 | — | — | — | |
| SaliencyBackbone=VGG16, Method Type=White box2021.11 | 30.3 | — | — | — | |
| SmoothGradBackbone=EfficientNet, Method Type=White box2021.11 | 29.9 | — | — | — | |
| Sobol signed (S+Ti)Backbone=VGG16, Method Type=Black box2021.11 | 29 | — | — | — | |
| SaliencyBackbone=VGG16, Access=White-box2022.06 | 28.6 | 12 | — | — | |
| Integ.-Grad.Backbone=VGG16, Access=White-box2022.06 | 27.6 | 11.4 | — | — | |
| LIMEBackbone=VGG16, Access=Black-box2022.06 | 27.3 | 25.8 | — | — | |
| Grad.-InputBackbone=VGG16, Access=White-box2022.06 | 27.2 | 11.6 | — | — | |
| Integ.-Grad.Backbone=ResNet50V2, Method Type=White box2021.11 | 26.4 | — | — | — | |
| Sobol signed (S+Ti)Backbone=ResNet50V2, Method Type=Black box2021.11 | 25.8 | — | — | — | |
| Integ.-Grad.Backbone=MobileNetV2, Access=White-box2022.06 | 25.8 | 9.6 | — | — | |
| SaliencyBackbone=MobileNetV2, Method Type=White box2021.11 | 25.3 | — | — | — | |
| Integ.-Grad.Backbone=EfficientNet, Access=White-box2022.06 | 24.8 | 7.8 | — | — | |
| SaliencyBackbone=MobileNetV2, Access=White-box2022.06 | 24.6 | 11.3 | — | — | |
| SmoothGradBackbone=MobileNetV2, Access=White-box2022.06 | 24.6 | 8.8 | — | — | |
| Guided-Backprop.Backbone=VGG16, Method Type=White box2021.11 | 24.2 | — | — | — | |
| VarGradBackbone=VGG16, Access=White-box2022.06 | 24.1 | 22.9 | — | — | |
| Integ.-Grad.Backbone=VGG16, Method Type=White box2021.11 | 23.7 | — | — | — | |
| Random BaselineBackbone=ResNet50V2, Method Type=White box2021.11 | 23.3 | — | — | — | |
| Grad.-InputBackbone=MobileNetV2, Access=White-box2022.06 | 23.1 | 11 | — | — | |
| SaliencyBackbone=EfficientNet, Method Type=White box2021.11 | 22.9 | — | — | — | |
| Guided-Backprop.Backbone=EfficientNet, Method Type=White box2021.11 | 22.9 | — | — | — | |
| DeconvNetBackbone=EfficientNet, Method Type=White box2021.11 | 22.9 | — | — | — | |
| SmoothGradBackbone=VGG16, Access=White-box2022.06 | 22.9 | 12.8 | — | — | |
| SaliencyBackbone=EfficientNet, Access=White-box2022.06 | 22.4 | 9.1 | — | — | |
| LIMEBackbone=EfficientNet, Access=Black-box2022.06 | 22.3 | 18.6 | — | — | |
| VarGradBackbone=EfficientNet, Access=White-box2022.06 | 22.2 | 22.4 | — | — | |
| DeconvNetBackbone=VGG16, Method Type=White box2021.11 | 22.1 | — | — | — | |
| Grad.-InputBackbone=EfficientNet, Access=White-box2022.06 | 22 | 8.4 | — | — | |
| Grad.-InputBackbone=VGG16, Method Type=White box2021.11 | 21.9 | — | — | — | |
| Sobol signed (S+Ti)Backbone=MobileNetV2, Method Type=Black box2021.11 | 21.1 | — | — | — | |
| Sobol signed (S+Ti)Backbone=EfficientNet, Method Type=Black box2021.11 | 20.4 | — | — | — | |
| Grad.-InputBackbone=ResNet50V2, Method Type=White box2021.11 | 19.4 | — | — | — | |
| SmoothGradBackbone=EfficientNet, Access=White-box2022.06 | 17.2 | 9.4 | — | — | |
| Random BaselineBackbone=VGG16, Method Type=White box2021.11 | 16.6 | — | — | — | |
| DeconvNetBackbone=MobileNetV2, Method Type=White box2021.11 | 16.6 | — | — | — | |
| Integ.-Grad.Backbone=MobileNetV2, Method Type=White box2021.11 | 16.6 | — | — | — | |
| OcclusionBackbone=ResNet50V2, Method Type=Black box2021.11 | 15.4 | — | — | — | |
| OcclusionBackbone=EfficientNet, Method Type=Black box2021.11 | 15.2 | — | — | — | |
| Integ.-Grad.Backbone=EfficientNet, Method Type=White box2021.11 | 14.3 | — | — | — | |
| Random BaselineBackbone=MobileNetV2, Method Type=White box2021.11 | 13.8 | — | — | — | |
| OcclusionBackbone=MobileNetV2, Method Type=Black box2021.11 | 13.5 | — | — | — |