Model Inversion Attack on CelebA (private) and FFHQ (public) (test)
77.27Attack AccuracyKEDMI + LOMMA
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| KEDMI + LOMMATarget Model=IR152, Base Attack=KEDMI, Evaluation Model=face.evoLve2023.04 | 77.27 | 24.4 | 1,292.8 | |
| KEDMI + LOMMATarget Model=face.evoLve, Base Attack=KEDMI, Evaluation Model=face.evoLve2023.04 | 73.2 | 21.33 | 1,321 | |
| GMI + LOMMATarget Model=IR152, Base Attack=GMI, Evaluation Model=face.evoLve2023.04 | 72 | 54.8 | 1,338.35 | |
| KEDMI + LOMTarget Model=face.evoLve, Base Attack=KEDMI, Evaluation Model=face.evoLve2023.04 | 69.73 | 17.86 | 1,379.73 | |
| KEDMI + LOMTarget Model=IR152, Base Attack=KEDMI, Evaluation Model=face.evoLve2023.04 | 67.73 | 14.86 | 1,325.28 | |
| KEDMI + MATarget Model=face.evoLve, Base Attack=KEDMI, Evaluation Model=face.evoLve2023.04 | 65.73 | 13.86 | 1,379.09 | |
| GMI + LOMMATarget Model=face.evoLve, Base Attack=GMI, Evaluation Model=face.evoLve2023.04 | 64.33 | 50.06 | 1,386.33 | |
| KEDMI + MATarget Model=IR152, Base Attack=KEDMI, Evaluation Model=face.evoLve2023.04 | 64.13 | 11.26 | 1,373.42 | |
| KEDMI + LOMMATarget Model=VGG16, Base Attack=KEDMI, Evaluation Model=face.evoLve2023.04 | 62.67 | 21.4 | 1,366.94 | |
| GMI + LOMMATarget Model=VGG16, Base Attack=GMI, Evaluation Model=face.evoLve2023.04 | 58.73 | 47.8 | 1,415.06 | |
| GMI + LOMTarget Model=IR152, Base Attack=GMI, Evaluation Model=face.evoLve2023.04 | 56 | 38.8 | 1,427.59 | |
| KEDMI + LOMTarget Model=VGG16, Base Attack=KEDMI, Evaluation Model=face.evoLve2023.04 | 55.07 | 13.8 | 1,438.72 | |
| KEDMITarget Model=IR152, Base Attack=KEDMI, Evaluation Model=face.evoLve2023.04 | 52.87 | — | 1,418.83 | |
| KEDMI + MATarget Model=VGG16, Base Attack=KEDMI, Evaluation Model=face.evoLve2023.04 | 52.07 | 10.8 | 1,428.77 | |
| KEDMITarget Model=face.evoLve, Base Attack=KEDMI, Evaluation Model=face.evoLve2023.04 | 51.87 | — | 1,440.19 | |
| GMI + MATarget Model=IR152, Base Attack=GMI, Evaluation Model=face.evoLve2023.04 | 50.8 | 33.6 | 1,462.92 | |
| GMI + LOMTarget Model=face.evoLve, Base Attack=GMI, Evaluation Model=face.evoLve2023.04 | 47.93 | 33.66 | 1,498.19 | |
| GMI + MATarget Model=face.evoLve, Base Attack=GMI, Evaluation Model=face.evoLve2023.04 | 46.07 | 31.8 | 1,500.1 | |
| GMI + LOMTarget Model=VGG16, Base Attack=GMI, Evaluation Model=face.evoLve2023.04 | 44.4 | 33.47 | 1,508.84 | |
| KEDMITarget Model=VGG16, Base Attack=KEDMI, Evaluation Model=face.evoLve2023.04 | 41.27 | — | 1,490.09 | |
| GMI + MATarget Model=VGG16, Base Attack=GMI, Evaluation Model=face.evoLve2023.04 | 34.93 | 24 | 1,547.93 | |
| GMITarget Model=IR152, Base Attack=GMI, Evaluation Model=face.evoLve2023.04 | 17.2 | — | 1,701.76 | |
| GMITarget Model=face.evoLve, Base Attack=GMI, Evaluation Model=face.evoLve2023.04 | 14.27 | — | 1,744.47 | |
| GMITarget Model=VGG16, Base Attack=GMI, Evaluation Model=face.evoLve2023.04 | 10.93 | — | 1,766.27 |