Table 1.
The open-source facial recognition DNNs used in the international forensic proficiency test.
| Name | Model | Training | Python library |
|---|---|---|---|
| DNN1 | VGG1680 | VGGFace80 | Keras |
| DNN2 | ResNet3481 | VGGFace80; Face Scrub dataset82 and images from the internet83 | Pytorch |
| DNN3 | ResNet5081 | VGGFace284 | Keras |
| DNN4 | ResNet5081 | VGGFace284 | Pytorch |
| DNN5 | ResNet5081 | MS-Celeb-1 M dataset85 fine-tuned on VGGFace284 | Pytorch |
| DNN6 | Se-ResNet5086 | VGGFace284 | Pytorch |
| DNN7 | Se-ResNet5086 | MS-Celeb-1 M dataset85 fine-tuned on VGGFace284 | Pytorch |
| DNN8 | ArcFace87 | CASIA88, VGGFace284, MS1MV287 and DeepGlint-Face89 | DeepFace90 |
| DNN9 | Facenet91 | VGGFace284 | DeepFace90 |
| DNN10 | Facenet51291 | VGGFace284 | DeepFace90 |