1
$\begingroup$

I'm building a face verification system using face embeddings (of 512 dimensions) from a pretrained Facenet model. For this, if I have some 4 to 5 images for a person, how can I successfully verify a new unseen images as either the same or different person?

I think this of one class classification (because of having single class) task, and googled it, but unable to find any reliable sources which suits well for this task.

Then, I've trained a SVM classifier with embeddings of a person and also with dummy embeddings of some 4 to 5 persons (so that having positive and negative classes). But it doesn't seem to work well.

Please suggest me with a ML algorithm/technique for this task, thank you.

$\endgroup$

1 Answer 1

1
$\begingroup$

I am assuming Facenet is an image classifier and it will give embedding for a face similar to other CNNs. If that is the case you don't need to train different classifiers, you can just remove the head of Facenet and initialize a small network on top of it and train an end to end network for better accuracies.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.