I wonder what image transformations would yield a different hash in the NeuralHash algorithm. Anyone know?
From Apple: “The neural network that generates the descriptor is trained through a self-supervised training scheme. Images are perturbed with transformations that keep them perceptually identical to the original, creating an original/perturbed pair.”
Not a pair. The number of possible perturbations of each example image in deep learning are enormous. This blithe quote implies that a hash code somehow can provide a flexible match against the billions of perturbations even a single image can undergo. It's a specious claim, and intended to cow anyone from disbelieving in the effectiveness of such a brute force match model.
From Apple: “The neural network that generates the descriptor is trained through a self-supervised training scheme. Images are perturbed with transformations that keep them perceptually identical to the original, creating an original/perturbed pair.”