Use image metadata to suggest clusters, i.e. identify that 50% of images use camera X, 50% use camera Y and that corresponds to label “cat” and label “dog”.
This is useful for checking that there aren't any hidden issues in the dataset, i.e. all pictures of cats are taken from a smartphone but dogs use a DSLR and so an AI might learn the difference between camera noise characteristics and not what's a cat.
Use image metadata to suggest clusters, i.e. identify that 50% of images use camera X, 50% use camera Y and that corresponds to label “cat” and label “dog”.
This is useful for checking that there aren't any hidden issues in the dataset, i.e. all pictures of cats are taken from a smartphone but dogs use a DSLR and so an AI might learn the difference between camera noise characteristics and not what's a cat.