I still can’t find him or her.
But I’m looking.
I’m not really looking for anyone specific. Just anyone I used to know.
Anyone who has ever walked through a shopping mall knows that there are so many different faces. I recall once that Vicky Loras told me she saw my Doppelgänger in Switzerland. I could swear that I saw Lisa Noble’s double in the Devonshire Mall in Windsor one day.
So, here’s my logic – such that it is.
I was inspired on this crusade using the website ThisPersonDoesNotExist. Created by Phillip Wang, it generates lifelike human faces from an algorithm. You can read all about it here and by following a few links followed by a few more links to get lots of details. Plus some interesting code to read, if you’re so inclined.
Abstract: We propose an alternative generator architecture for generative adversarial networks, borrowing from style transfer literature. The new architecture leads to an automatically learned, unsupervised separation of high-level attributes (e.g., pose and identity when trained on human faces) and stochastic variation in the generated images (e.g., freckles, hair), and it enables intuitive, scale-specific control of the synthesis. The new generator improves the state-of-the-art in terms of traditional distribution quality metrics, leads to demonstrably better interpolation properties, and also better disentangles the latent factors of variation. To quantify interpolation quality and disentanglement, we propose two new, automated methods that are applicable to any generator architecture. Finally, we introduce a new, highly varied and high-quality dataset of human faces.
She doesn’t exist
So, all weekend, when I felt the urge, I kept whacking CTRL-R to get a new face. I’m here to report that I haven’t found someone I used to know yet.
I can see some interesting uses for this in the classroom. For those higher end computer science students, the reading is interesting just to see what is possible.
In terms of basic media literacy though, it presents a concrete example as to how things can be created from nothing more than a few electronic bits (and some pretty awesome programming). It also poses an interesting inquiry to generate a face and then very closely analyse it. Are there clues that would let you know that it’s not a real photo?