Model Inversion Defense on CelebA 64x64
91.16AccuracyNoDef
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| NoDefAttack=GMI, Architecture=IR152, Resolution=64x64, Private Dataset=CelebA, Public Dataset=CelebA2024.09 | 91.16 | 32.4 | 1,587.28 | — | |
| NoDefAttack=KedMI, Architecture=IR152, Resolution=64x64, Private Dataset=CelebA, Public Dataset=CelebA2024.09 | 91.16 | 78.93 | 1,262.44 | — | |
| NoDefAttack=LOMMA + GMI, Architecture=IR152, Resolution=64x64, Private Dataset=CelebA, Public Dataset=CelebA2024.09 | 91.16 | 80.93 | 1,253.03 | — | |
| NoDefAttack=LOMMA + KedMI, Architecture=IR152, Resolution=64x64, Private Dataset=CelebA, Public Dataset=CelebA2024.09 | 91.16 | 90.87 | 1,116.9 | — | |
| NoDefAttack=PLGMI, Architecture=IR152, Resolution=64x64, Private Dataset=CelebA, Public Dataset=CelebA2024.09 | 91.16 | 99.47 | 1,021.42 | — | |
| NoDefAttack=GMI, Target Model (T)=FaceNet64, Resolution=64x642024.09 | 88.5 | 29.6 | 1,607.86 | — | |
| NoDefAttack=KedMI, Target Model (T)=FaceNet64, Resolution=64x642024.09 | 88.5 | 81.67 | 1,270.71 | — | |
| NoDefAttack=LOMMA + GMI, Target Model (T)=FaceNet64, Resolution=64x642024.09 | 88.5 | 83.33 | 1,259.61 | — | |
| NoDefAttack=LOMMA + KedMI, Target Model (T)=FaceNet64, Resolution=64x642024.09 | 88.5 | 90.87 | 1,116.9 | — | |
| NoDefAttack=PLGMI, Target Model (T)=FaceNet64, Resolution=64x642024.09 | 88.5 | 99.47 | 1,091.51 | — | |
| NoDefAttack=GMI, Target Model=FaceNet64, Private Dataset=CelebA, Public Dataset=CelebA2024.09 | 88.5 | 29.6 | 1,607.86 | — | |
| NoDefAttack=KedMI, Target Model=FaceNet64, Private Dataset=CelebA, Public Dataset=CelebA2024.09 | 88.5 | 81.67 | 1,270.71 | — | |
| NoDefAttack=LOMMA + GMI, Target Model=FaceNet64, Private Dataset=CelebA, Public Dataset=CelebA2024.09 | 88.5 | 83.33 | 1,259.61 | — | |
| NoDefAttack=LOMMA + KedMI, Target Model=FaceNet64, Private Dataset=CelebA, Public Dataset=CelebA2024.09 | 88.5 | 90.87 | 1,116.9 | — | |
| MIDREAttack=GMI, Target Model=FaceNet64, Private Dataset=CelebA, Public Dataset=CelebA2024.09 | 85.74 | 7.47 | 1,898.29 | 8.02 | |
| MIDREAttack=KedMI, Target Model=FaceNet64, Private Dataset=CelebA, Public Dataset=CelebA2024.09 | 85.74 | 42.93 | 1,512.52 | 14.04 | |
| MIDREAttack=LOMMA + GMI, Target Model=FaceNet64, Private Dataset=CelebA, Public Dataset=CelebA2024.09 | 85.74 | 43.33 | 1,550.77 | 14.49 | |
| MIDREAttack=LOMMA + KedMI, Target Model=FaceNet64, Private Dataset=CelebA, Public Dataset=CelebA2024.09 | 85.74 | 58.07 | 1,386.67 | 11.88 | |
| NoDefAttack=LOMMA + GMI, Target Model (T)=VGG16, Resolution=64x64, Private Dataset (Dpriv)=CelebA, Public Dataset (Dpub)=CelebA2024.09 | 85.74 | 53.64 | — | — | |
| NoDefAttack=LOMMA + KedMI, Target Model (T)=VGG16, Resolution=64x64, Private Dataset (Dpriv)=CelebA, Public Dataset (Dpub)=CelebA2024.09 | 85.74 | 72.96 | — | — | |
| NoDefAttack=PLGMI, Target Model (T)=VGG16, Resolution=64x64, Private Dataset (Dpriv)=CelebA, Public Dataset (Dpub)=CelebA2024.09 | 85.74 | 71 | — | — | |
| MIDREAttack=GMI, Architecture=IR152, Resolution=64x64, Private Dataset=CelebA, Public Dataset=CelebA2024.09 | 84.91 | 7.87 | 1,888.47 | — | |
| MIDREAttack=KedMI, Architecture=IR152, Resolution=64x64, Private Dataset=CelebA, Public Dataset=CelebA2024.09 | 84.91 | 40.07 | 1,548.16 | — | |
| MIDREAttack=LOMMA + GMI, Architecture=IR152, Resolution=64x64, Private Dataset=CelebA, Public Dataset=CelebA2024.09 | 84.91 | 40.93 | 1,559.88 | — | |
| MIDREAttack=LOMMA + KedMI, Architecture=IR152, Resolution=64x64, Private Dataset=CelebA, Public Dataset=CelebA2024.09 | 84.91 | 52.13 | 1,481.7 | — | |
| MIDREAttack=PLGMI, Architecture=IR152, Resolution=64x64, Private Dataset=CelebA, Public Dataset=CelebA2024.09 | 84.91 | 77.4 | 1,470.46 | — | |
| TL-DMIAttack=GMI, Target Model=FaceNet64, Private Dataset=CelebA, Public Dataset=CelebA2024.09 | 83.41 | 15.73 | 1,752 | 2.72 | |
| TL-DMIAttack=KedMI, Target Model=FaceNet64, Private Dataset=CelebA, Public Dataset=CelebA2024.09 | 83.41 | 73.4 | 1,265 | 1.62 | |
| TL-DMIAttack=LOMMA + GMI, Target Model=FaceNet64, Private Dataset=CelebA, Public Dataset=CelebA2024.09 | 83.41 | 43.67 | 1,616 | 7.79 | |
| TL-DMIAttack=LOMMA + KedMI, Target Model=FaceNet64, Private Dataset=CelebA, Public Dataset=CelebA2024.09 | 83.41 | 79.6 | 1,345 | 2.21 | |
| MIDREAttack=GMI, Target Model (T)=FaceNet64, Resolution=64x642024.09 | 81.56 | 6.73 | 1,908.19 | — | |
| MIDREAttack=KedMI, Target Model (T)=FaceNet64, Resolution=64x642024.09 | 81.56 | 36.33 | 1,545.93 | — | |
| MIDREAttack=LOMMA + GMI, Target Model (T)=FaceNet64, Resolution=64x642024.09 | 81.56 | 37.6 | 1,570.85 | — | |
| MIDREAttack=LOMMA + KedMI, Target Model (T)=FaceNet64, Resolution=64x642024.09 | 81.56 | 54.33 | 1,456.84 | — | |
| MIDREAttack=PLGMI, Target Model (T)=FaceNet64, Resolution=64x642024.09 | 81.56 | 75 | 1,509.78 | — | |
| NLSAttack=LOMMA + GMI, Target Model (T)=VGG16, Resolution=64x64, Private Dataset (Dpriv)=CelebA, Public Dataset (Dpub)=CelebA2024.09 | 80.02 | 39.16 | — | 2.53 | |
| NLSAttack=LOMMA + KedMI, Target Model (T)=VGG16, Resolution=64x64, Private Dataset (Dpriv)=CelebA, Public Dataset (Dpub)=CelebA2024.09 | 80.02 | 63.6 | — | 1.64 | |
| NLSAttack=PLGMI, Target Model (T)=VGG16, Resolution=64x64, Private Dataset (Dpriv)=CelebA, Public Dataset (Dpub)=CelebA2024.09 | 80.02 | 72 | — | -0.17 | |
| MIDREAttack=LOMMA + GMI, Target Model (T)=VGG16, Resolution=64x64, Private Dataset (Dpriv)=CelebA, Public Dataset (Dpub)=CelebA2024.09 | 79.85 | 26.62 | — | 4.59 | |
| MIDREAttack=LOMMA + KedMI, Target Model (T)=VGG16, Resolution=64x64, Private Dataset (Dpriv)=CelebA, Public Dataset (Dpub)=CelebA2024.09 | 79.85 | 41.82 | — | 5.29 | |
| MIDREAttack=PLGMI, Target Model (T)=VGG16, Resolution=64x64, Private Dataset (Dpriv)=CelebA, Public Dataset (Dpub)=CelebA2024.09 | 79.85 | 66.6 | — | 0.75 |