Model Inversion Attack Defense on CelebA (test)
0.645SSIMSL
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| SLDescription=Standard Learning baseline without privacy protection2023.04 | 0.645 | 16.25 | 0.933 | |
| GNTechnique=Gaussian Noise2023.04 | 0.457 | 15.47 | 0.663 | |
| LPTechnique=Loss-pass filter, Radius=0.052023.04 | 0.45 | 14.71 | 0.614 | |
| GN+Technique=Gaussian Noise, Enhanced by row permutation=true, Permutation matrix shape=197 x 1972023.04 | 0.167 | 8.068 | 0.187 | |
| BSTechnique=Batch Shuffle2023.04 | 0.148 | 8.18 | 0.176 | |
| BS+Technique=Batch Shuffle, Enhanced by row permutation=true, Permutation matrix shape=197 x 1972023.04 | 0.143 | 7.613 | 0.167 | |
| LP+Technique=Loss-pass filter, Radius=0.05, Enhanced by row permutation=true, Permutation matrix shape=197 x 1972023.04 | 0.11 | 5.687 | 0.154 |