Raw LLM Responses

I took a computer vision course in my last semester at uni and our professor–an extremely well-respected figure in his field with a shitload (>200) of publications–set aside half a class session to discuss why facial recognition software is useless garbage that should stay in the research sector. First of all, there's a single application for it and that's surveillance and monitoring. And second, no one has a facial recog package that can guarantee 100% true positive. Given that the only application kinda needs that guaranteed accuracy, you and your conscience are better off finding something more productive to work on. Edit for the iPhone users: yeah faceID seems like a great example of facial recognition software right up until you learn it's not facial recognition. Modern FR software relies on decomposing the covariance matrices of training data into basis vectors that tell you the "shape" of pixel data that contain that face. Matching involves (among other things) decomposing your test image the same way and comparing the basis vectors. This is the facial recognition approach you would use on footage from security cameras, phones, or anything else that's "just a camera". While I don't know the exact details of FaceID, I do know that the technique is A: not the same at all, and B: impossible without specialized hardware. The iphone projects a bunch of IR dots onto your face, and using their deformation constructs the 3D representation of your face to compare against later. Nothing to do with "the shape of the data from an image containing your face"