Doubts about the Siamese Network with Triplet Loss Function

Wait since the input of the model are three images how do you actually perform image verification on single images after training the model? Someone please help me with this I’ve been trying to find the answer countless times but still can’t seem to succeed.

Hey @James_Nathan_O,
Apologies for the delayed response. The triplet loss is just a cost function to train our Siamese network, and even if we consider the triplet loss, the network is still common for obtaining the embeddings of A, P and N images, and in order to obtain the embedding of any single image, we don’t need the other 2 images.

So, if we consider the example of face verification, i.e., match whether a person is who he/she says, once the network is trained, we can simply cross reference the embeddings for his/her current image (which is just clicked) to the one stored in the database (which would be retrieved with the help of his/her ID). I hope this helps.

Regards,
Elemento