Model Inversion Defense on CelebA 64x64 (test)
86.9AccuracyNoDef
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| NoDefAttack=LOMMA + GMI, Target Model (T)=VGG16, Private Dataset (Dpriv)=CelebA, Public Dataset (Dpub)=CelebA2024.09 | 86.9 | 74.53 | — | 1,312.93 | |
| NoDefAttack=LOMMA + KedMI, Target Model (T)=VGG16, Private Dataset (Dpriv)=CelebA, Public Dataset (Dpub)=CelebA2024.09 | 86.9 | 81.8 | — | 1,211.45 | |
| NoDefAttack=PLGMI, Target Model (T)=VGG16, Private Dataset (Dpriv)=CelebA, Public Dataset (Dpub)=CelebA2024.09 | 86.9 | 97.47 | — | 1,149.67 | |
| NoDefAttack=GMI, Target Model (T)=VGG16, Private Dataset (Dpriv)=CelebA, Public Dataset (Dpub)=CelebA2024.09 | 86.9 | 20.07 | — | 1,679.18 | |
| NoDefAttack=KedMI, Target Model (T)=VGG16, Private Dataset (Dpriv)=CelebA, Public Dataset (Dpub)=CelebA2024.09 | 86.9 | 78.47 | — | 1,289.46 | |
| NoDefAttack=BREPMI, Target Model (T)=VGG16, Private Dataset (Dpriv)=CelebA, Public Dataset (Dpub)=CelebA2024.09 | 86.9 | 57.4 | — | 1,376.94 | |
| BiDOAttack=LOMMA + GMI, Target Model (T)=VGG16, Private Dataset (Dpriv)=CelebA, Public Dataset (Dpub)=CelebA2024.09 | 79.85 | 53.73 | 2.95 | 1,422.75 | |
| MIDREAttack=LOMMA + GMI, Target Model (T)=VGG16, Private Dataset (Dpriv)=CelebA, Public Dataset (Dpub)=CelebA2024.09 | 79.85 | 31.93 | 6.04 | 1,590.12 | |
| BiDOAttack=LOMMA + KedMI, Target Model (T)=VGG16, Private Dataset (Dpriv)=CelebA, Public Dataset (Dpub)=CelebA2024.09 | 79.85 | 63 | 2.67 | 1,345.94 | |
| MIDREAttack=LOMMA + KedMI, Target Model (T)=VGG16, Private Dataset (Dpriv)=CelebA, Public Dataset (Dpub)=CelebA2024.09 | 79.85 | 43.07 | 5.49 | 1,503.89 | |
| BiDOAttack=PLGMI, Target Model (T)=VGG16, Private Dataset (Dpriv)=CelebA, Public Dataset (Dpub)=CelebA2024.09 | 79.85 | 92.4 | 0.72 | 1,228.36 | |
| MIDREAttack=PLGMI, Target Model (T)=VGG16, Private Dataset (Dpriv)=CelebA, Public Dataset (Dpub)=CelebA2024.09 | 79.85 | 66.6 | 4.38 | 1,475.76 | |
| BiDOAttack=GMI, Target Model (T)=VGG16, Private Dataset (Dpriv)=CelebA, Public Dataset (Dpub)=CelebA2024.09 | 79.85 | 6.13 | 1.98 | 1,927.11 | |
| MIDREAttack=GMI, Target Model (T)=VGG16, Private Dataset (Dpriv)=CelebA, Public Dataset (Dpub)=CelebA2024.09 | 79.85 | 3.2 | 2.39 | 2,020.49 | |
| BiDOAttack=KedMI, Target Model (T)=VGG16, Private Dataset (Dpriv)=CelebA, Public Dataset (Dpub)=CelebA2024.09 | 79.85 | 43.53 | 4.96 | 1,494.35 | |
| MIDREAttack=KedMI, Target Model (T)=VGG16, Private Dataset (Dpriv)=CelebA, Public Dataset (Dpub)=CelebA2024.09 | 79.85 | 34.73 | 6.2 | 1,620.66 | |
| BiDOAttack=BREPMI, Target Model (T)=VGG16, Private Dataset (Dpriv)=CelebA, Public Dataset (Dpub)=CelebA2024.09 | 79.85 | 37.4 | 2.84 | 1,500.45 | |
| MIDREAttack=BREPMI, Target Model (T)=VGG16, Private Dataset (Dpriv)=CelebA, Public Dataset (Dpub)=CelebA2024.09 | 79.85 | 21.73 | 5.06 | 1,611.78 | |
| MIDAttack=LOMMA + GMI, Target Model (T)=VGG16, Private Dataset (Dpriv)=CelebA, Public Dataset (Dpub)=CelebA2024.09 | 79.16 | 54.53 | 2.58 | 1,348.21 | |
| MIDAttack=LOMMA + KedMI, Target Model (T)=VGG16, Private Dataset (Dpriv)=CelebA, Public Dataset (Dpub)=CelebA2024.09 | 79.16 | 67.2 | 1.89 | 1,249.18 | |
| MIDAttack=PLGMI, Target Model (T)=VGG16, Private Dataset (Dpriv)=CelebA, Public Dataset (Dpub)=CelebA2024.09 | 79.16 | 93 | 0.58 | 1,111.61 | |
| MIDAttack=GMI, Target Model (T)=VGG16, Private Dataset (Dpriv)=CelebA, Public Dataset (Dpub)=CelebA2024.09 | 79.16 | 20.93 | -0.11 | 1,698.5 | |
| MIDAttack=KedMI, Target Model (T)=VGG16, Private Dataset (Dpriv)=CelebA, Public Dataset (Dpub)=CelebA2024.09 | 79.16 | 53.33 | 3.25 | 1,364.02 | |
| MIDAttack=BREPMI, Target Model (T)=VGG16, Private Dataset (Dpriv)=CelebA, Public Dataset (Dpub)=CelebA2024.09 | 79.16 | 39.2 | 2.35 | 1,458.61 |