Model Inversion Attack on VGGFace
97.96Top-1 AccuracyPPA
Evaluation results
| Method | Links | ||
|---|---|---|---|
| PPATarget Model=VGG16, Image Prior=CelebA2025.03 | 97.96 | 100 | |
| Mirror-wTarget Model=VGG16, Image Prior=CelebA2025.03 | 81.63 | 89.8 | |
| SMILE (Swin-T)Target Model=VGG16, Image Prior=FFHQ, Surrogate Model=Swin-T2025.03 | 80.95 | 88.44 | |
| SMILE (InceptionV3)Target Model=VGG16, Image Prior=CelebA, Surrogate Model=InceptionV32025.03 | 72.11 | 87.76 | |
| SMILE (Swin-T)Target Model=VGG16, Image Prior=CelebA, Surrogate Model=Swin-T2025.03 | 71.9 | 83.53 | |
| RLBMITarget Model=VGG16, Image Prior=CelebA2025.03 | 71.43 | 83.67 | |
| SMILE (InceptionV1*)Target Model=VGG16, Image Prior=CelebA, Surrogate Model=InceptionV1, Initialization=Pre-trained2025.03 | 71.43 | 89.12 | |
| SMILE (MobileNetV2)Target Model=VGG16, Image Prior=CelebA, Surrogate Model=MobileNetV22025.03 | 71.43 | 82.99 | |
| SMILE (EfficientNetB0)Target Model=VGG16, Image Prior=CelebA, Surrogate Model=EfficientNetB02025.03 | 70.75 | 84.35 | |
| SMILE (ResNet50*)Target Model=VGG16, Image Prior=CelebA, Surrogate Model=ResNet50, Initialization=Pre-trained2025.03 | 70.07 | 87.76 | |
| Mirror-bTarget Model=VGG16, Image Prior=CelebA2025.03 | 59.18 | 73.47 |