Well, I think the new method is actually useful in a lot of cases and the old method has a totally different use case (i.e. finding more occurrences of the exac

Well, I think the new method is actually useful in a lot of cases and the old method has a totally different use case (i.e. finding more occurrences of the exact same image). The ML approach is good for finding similar images that aren't the exact same image. Which, honestly, seems like a much more common use case. The old fingerprinting method is pretty much just used for investigative purposes like finding out where an image originated from. Something I liked to use it for, but I doubt is a super common thing for your average user.