Model Inversion Attack on CelebA (private) FFHQ (public) on VGG16 (test)
86.33Top-5 Attack AccuracyGMI + LOMMA
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| GMI + LOMMATarget Model=VGG16, Evaluation Model=face.evoLve, Public Dataset=FFHQ, Private Dataset=CelebA2023.04 | 86.33 | 60.67 | 35.59 | |
| KEDMI + LOMMATarget Model=VGG16, Evaluation Model=face.evoLve, Public Dataset=FFHQ, Private Dataset=CelebA2023.04 | 85.33 | 11.33 | 40.26 | |
| KEDMI + LOMTarget Model=VGG16, Evaluation Model=face.evoLve, Public Dataset=FFHQ, Private Dataset=CelebA2023.04 | 81.67 | 7.67 | 43.76 | |
| KEDMI + MATarget Model=VGG16, Evaluation Model=face.evoLve, Public Dataset=FFHQ, Private Dataset=CelebA2023.04 | 80.33 | 6.33 | 35.02 | |
| KEDMITarget Model=VGG16, Evaluation Model=face.evoLve, Public Dataset=FFHQ, Private Dataset=CelebA2023.04 | 74 | — | 36.18 | |
| GMI + LOMTarget Model=VGG16, Evaluation Model=face.evoLve, Public Dataset=FFHQ, Private Dataset=CelebA2023.04 | 70.67 | 45 | 42.6 | |
| GMI + MATarget Model=VGG16, Evaluation Model=face.evoLve, Public Dataset=FFHQ, Private Dataset=CelebA2023.04 | 62.33 | 36.67 | 36.04 | |
| GMITarget Model=VGG16, Evaluation Model=face.evoLve, Public Dataset=FFHQ, Private Dataset=CelebA2023.04 | 25.67 | — | 53.17 |