Feature Inversion on ImageNet
43.2SSIMAdjoint Inversion
Evaluation results
| Method | Links | ||
|---|---|---|---|
| Adjoint InversionBackbone=ConvNeXt-Base2026.04 | 43.2 | 53.5 | |
| Adjoint InversionBackbone=ResNet-502026.04 | 41.7 | 43 | |
| Adjoint InversionBackbone=DenseNet-1212026.04 | 38.5 | 45.8 | |
| Adjoint InversionBackbone=ResNet-182026.04 | 34.6 | 57.7 | |
| UpConvNetBackbone=ResNet-182026.04 | 18.5 | 92.1 | |
| UpConvNetBackbone=ResNet-502026.04 | 15.9 | 89.3 | |
| UpConvNetBackbone=DenseNet-1212026.04 | 14.5 | 92.2 | |
| UpConvNetBackbone=ConvNeXt-Base2026.04 | 12 | 89 | |
| RobustBackbone=ConvNeXt-Base2026.04 | 9.6 | 101.6 | |
| MahendranBackbone=ConvNeXt-Base2026.04 | 8.7 | 104.5 | |
| RobustBackbone=DenseNet-1212026.04 | 8.7 | 116.8 | |
| RobustBackbone=ResNet-502026.04 | 7.6 | 105.2 | |
| MahendranBackbone=ResNet-502026.04 | 7.5 | 99 | |
| MahendranBackbone=DenseNet-1212026.04 | 6.9 | 108.8 | |
| RobustBackbone=ResNet-182026.04 | 6.5 | 113.6 | |
| MahendranBackbone=ResNet-182026.04 | 6.2 | 107 | |
| Vanilla GradientBackbone=ResNet-502026.04 | 4.8 | 109.5 | |
| Vanilla GradientBackbone=DenseNet-1212026.04 | 4.3 | 107.9 | |
| Vanilla GradientBackbone=ConvNeXt-Base2026.04 | 3.7 | 108.8 | |
| Vanilla GradientBackbone=ResNet-182026.04 | 2.5 | 119 |