Model Inversion Attack on CelebA
93.93Attack AccLOKT
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| LOKTTarget Model (T)=FaceNet64, Public Dataset (Dpub)=CelebA, Design=Sen2023.10 | 93.93 | 1,181.72 | |
| LOKTTarget Model (T)=FaceNet64, Public Dataset (Dpub)=CelebA, Design=S2023.10 | 92.8 | 1,207.25 | |
| LOKTTarget Model (T)=IR152, Public Dataset (Dpub)=CelebA, Design=Sen2023.10 | 92.13 | 1,206.78 | |
| LOKTTarget Model (T)=IR152, Public Dataset (Dpub)=CelebA, Design=S2023.10 | 89.8 | 1,220 | |
| LOKTTarget Model (T)=VGG16, Public Dataset (Dpub)=CelebA, Design=Sen2023.10 | 87.27 | 1,246.71 | |
| LOKTTarget Model (T)=VGG16, Public Dataset (Dpub)=CelebA, Design=S2023.10 | 85.6 | 1,252.09 | |
| LOKTTarget Model (T)=FaceNet64, Public Dataset (Dpub)=CelebA, Design=CoD2023.10 | 81 | 1,298.63 | |
| BREPMITarget Model (T)=FaceNet64, Public Dataset (Dpub)=CelebA2023.10 | 73.93 | 1,284.41 | |
| LOKTTarget Model (T)=IR152, Public Dataset (Dpub)=CelebA, Design=CoD2023.10 | 72.07 | 1,358.94 | |
| BREPMITarget Model (T)=IR152, Public Dataset (Dpub)=CelebA2023.10 | 71.47 | 1,277.23 | |
| LOKTTarget Model (T)=VGG16, Public Dataset (Dpub)=CelebA, Design=CoD2023.10 | 71.33 | 1,364.47 | |
| LOKTTarget Model (T)=FaceNet64, Public Dataset (Dpub)=FFHQ, Design=Sen2023.10 | 62.07 | 1,428.04 | |
| LOKTTarget Model (T)=FaceNet64, Public Dataset (Dpub)=FFHQ, Design=S2023.10 | 59.13 | 1,437.86 | |
| BREPMITarget Model (T)=VGG16, Public Dataset (Dpub)=CelebA2023.10 | 57.4 | 1,376.94 | |
| LOKTTarget Model (T)=FaceNet64, Public Dataset (Dpub)=FFHQ, Design=CoD2023.10 | 43.27 | 1,516.18 | |
| MIDREAttack=Standard inversion attack2024.09 | 43.07 | — | |
| BREPMITarget Model (T)=FaceNet64, Public Dataset (Dpub)=FFHQ2023.10 | 43 | 1,470.55 | |
| MIDRE (Adapt.Att)Attack=Adaptive attack (attacker knows masking portions)2024.09 | 38.53 | — |