Discussion
For the ten thousandth time, facial recognition software does not graphically present faces and matchpoints as it runs.
Sugar, graphics are computationally *expensive*.
Those apps don't work by sequentially comparing every face in their dataset to the target face, but even if they *did*, they would not waste precious computational power showing rejects on a fucking screen.
Neighbor search in high-dimensional space is fucking rocket-science, and there is no current definitive answer to it.
But I can assure you, whatever a viable algorithm might be, it is not a) searching the dataset sequentially, or b) graphically rendering each face it is processing w/its matchpoints.
Nobody believes me, but the assertion is easy to prove. Go to the web and look up high dimensional space search.
Within two paragraphs of any article you find, you'll be deep in mathematical post-doc territory.
There are some viable semi-working solutions, but they all operate in *extremely* narrow and constrained contexts, and their error epsilon is--technical term here--*shite*.
We don't even have *fingerprint* search space fully worked out, and we have way better specimens and several dimensions which number <100 or even <10 variants to guide us deeper into the search.
And, no, *fingerprint* searches also don't a) sequentially evaluate within the dataset or b) draw fingerprints on the screen.
@GeePawHill TV show's designer: "...but progress bars and spinners are SO boring!..."
Designer continues by designing a DNA search that displays "artist's impression of a twisting strand of DNA [of the wrong chirality]", and shows it performing linear database search, for some techno thriller show about some pandemic.
Meanwhile, actual screen output of a bioinformatics pipeline:
https://mstdn.science/@dryak/112322539881161036
@GeePawHill pretty sure rockets only operate in, like, three or four dimensions and that seems hard enough already
Install bonfire.cafe
Get the full app experience
