Evading state of the art face verification system such as VGG-Face, Google FaceNet, OpenFace, Facebook DeepFace, DeepID, ArcFace and Dlib.
- [] Find test accuracy on the base pre-trained models
- Take a library of faces in a folder
- Find embedding of all the images and store in a database/ pickle file (use represent function from DeepFace, note to use the "model" parameter in it to avoid loading)
- [] Find closest embeddings and make pairs.
- [] Perform adversarial attack on the same.