List your company
ResearchFormerly WizWand

Model Inversion on CelebA (test)

93.2Attack AccuracyKEDMI + LOMMA
-3.384821.690146.76571.8399Mar 3, 2022May 8, 2022Jul 13, 2022Sep 17, 2022Nov 22, 2022Jan 27, 2023Apr 4, 2023
Updated 8mo ago

Evaluation results

MethodLinks
KEDMI + LOMMATarget Model=face.evoLve, Modification=Logit Maximization + Model Augmentation2023.0493.211.81,154.32
KEDMI + LOMMATarget Model=IR152, Modification=Logit Maximization + Model Augmentation2023.0492.9312.41,138.62
KEDMI + LOMTarget Model=face.evoLve, Modification=Logit Maximization2023.0492.5311.131,183.76
KEDMI + LOMTarget Model=IR152, Modification=Logit Maximization2023.0492.4711.941,168.55
KEDMI + LOMMATarget Model=VGG16, Modification=Logit Maximization + Model Augmentation2023.0490.2716.271,147.41
KEDMI + LOMTarget Model=VGG16, Modification=Logit Maximization2023.0489.0715.071,218.46
KEDMI + MATarget Model=face.evoLve, Modification=Model Augmentation2023.0485.073.671,222.02
KEDMI + MATarget Model=IR152, Modification=Model Augmentation2023.0484.734.21,220.23
KED-MIModel Architecture=IR152, Threat Model=Whitebox2022.0383——
GMI + LOMMATarget Model=IR152, Modification=Logit Maximization + Model Augmentation2023.0482.451.81,254.32
GMI + LOMMATarget Model=face.evoLve, Modification=Logit Maximization + Model Augmentation2023.0482.3355.261,257.5
KED-MIModel Architecture=FaceNet64, Threat Model=Whitebox2022.0382——
KEDMI + MATarget Model=VGG16, Modification=Model Augmentation2023.048281,248.33
KEDMITarget Model=face.evoLve2023.0481.4—1,248.32
KEDMITarget Model=IR1522023.0480.53—1,247.28
GMI + LOMTarget Model=IR152, Modification=Logit Maximization2023.0478.5347.931,289.62
GMI + LOMMATarget Model=VGG16, Modification=Logit Maximization + Model Augmentation2023.0477.658.531,296.26
BREP-MIModel Architecture=FaceNet64, Threat Model=Label-only2022.0375.67——
GMI + MATarget Model=face.evoLve, Modification=Model Augmentation2023.0474.1347.061,352.25
KEDMITarget Model=VGG162023.0474—1,289.88
BREP-MIModel Architecture=IR152, Threat Model=Label-only2022.0372——
GMI + LOMTarget Model=VGG16, Modification=Logit Maximization2023.0469.6750.61,363.81
KED-MIModel Architecture=VGG16, Threat Model=Whitebox2022.0369——
BREP-MIModel Architecture=VGG16, Threat Model=Label-only2022.0363.33——
GMI + LOMTarget Model=face.evoLve, Modification=Logit Maximization2023.0461.6734.61,405.35
GMI + MATarget Model=IR152, Modification=Model Augmentation2023.0461.230.61,389.99
GMI + MATarget Model=VGG16, Modification=Model Augmentation2023.0451.7332.661,467.68
GMIModel Architecture=FaceNet64, Threat Model=Whitebox2022.0332——
GMITarget Model=IR1522023.0430.6—1,609.29
GMITarget Model=face.evoLve2023.0427.07—1,635.87
GMIModel Architecture=IR152, Threat Model=Whitebox2022.0326——
GMITarget Model=VGG162023.0419.07—1,715.6
GMIModel Architecture=VGG16, Threat Model=Whitebox2022.0315——
LB-MIModel Architecture=FaceNet64, Threat Model=Blackbox2022.031.67——
LB-MIModel Architecture=VGG16, Threat Model=Blackbox2022.031.33——
LB-MIModel Architecture=IR152, Threat Model=Blackbox2022.030.33——