First Project Help

Hi guys, I just started my first project after finishing the deep learning specialization. I am using OpenCV’s YuNet model to detect and crop out faces on some images and OpenCV’s SFace model to generate and compare encodings. When I input 2 images that are very obviously different people I am getting a high cosine similarity score. This is strange considering that SFace scored 99.60% on Faces in the Wild which has some much harder test examples. Is my issue because of the varying image quality and size, or is it likely an error in my code somewhere.

image

I don’t know a lot about this model, but some initial searches says that it is a face detector, not so much as a reliable identifier of individuals.

Its claim to fame is that that is runs very quickly.

It does tasks like this:
image

Reference: