Feature Inversion on CUB-200
54.8SSIMAdjoint Inversion
Evaluation results
| Method | Links | ||
|---|---|---|---|
| Adjoint InversionBackbone=ResNet-502026.04 | 54.8 | 52.9 | |
| Adjoint InversionBackbone=ConvNeXt-Base2026.04 | 47.7 | 64.6 | |
| Adjoint InversionBackbone=ResNet-182026.04 | 44.2 | 64.4 | |
| UpConvNetBackbone=ResNet-502026.04 | 43.5 | 64.1 | |
| Adjoint InversionBackbone=DenseNet-1212026.04 | 40.7 | 61.5 | |
| UpConvNetBackbone=DenseNet-1212026.04 | 38.5 | 71.2 | |
| UpConvNetBackbone=ResNet-182026.04 | 36.4 | 76.8 | |
| UpConvNetBackbone=ConvNeXt-Base2026.04 | 36.2 | 87 | |
| MahendranBackbone=ConvNeXt-Base2026.04 | 12.5 | 95.1 | |
| MahendranBackbone=ResNet-502026.04 | 10.5 | 118.6 | |
| RobustBackbone=ConvNeXt-Base2026.04 | 8.9 | 105.8 | |
| MahendranBackbone=ResNet-182026.04 | 8.4 | 133.2 | |
| RobustBackbone=ResNet-502026.04 | 7.1 | 121 | |
| MahendranBackbone=DenseNet-1212026.04 | 6.2 | 127.3 | |
| Vanilla GradientBackbone=ConvNeXt-Base2026.04 | 6.1 | 134.1 | |
| Vanilla GradientBackbone=ResNet-502026.04 | 5 | 129.1 | |
| Vanilla GradientBackbone=DenseNet-1212026.04 | 4.9 | 133.9 | |
| RobustBackbone=DenseNet-1212026.04 | 4.9 | 131.5 | |
| RobustBackbone=ResNet-182026.04 | 4.7 | 132.1 | |
| Vanilla GradientBackbone=ResNet-182026.04 | 3.1 | 140.4 |