A live demo of calculus-ego, my open-source pipeline for person-level visual content analysis. Upload a photo — or go live — and the model detects every person and labels them (gender, age, behaviour, activity, body display, location, accessories and more), drawn right on top of each one. Everything runs on a private GPU server on my LAN and is never stored.
Upload a photo to see every person labelled.
Runs on a private LAN GPU server · Images are not stored
The original single-purpose demo: place a face in the frame and the model estimates a facial-attractiveness score from 1 to 10. Same private GPU server, nothing stored — just the number.
Run the demo to see an estimate.
Runs on a private LAN GPU server · Images are not stored
The same idea for words. verba-ego takes a social-media bio, splits it into the fragments its author wrote — one per line — and labels each on sense, reference and attribution, plus commercial for the bio as a whole. Nothing is generated: every candidate label is scored by log-likelihood, so each one leaves a probability behind. Same private GPU server, same model, nothing stored.
Write a bio — one line per fragment — and press Classify.
Runs on a private LAN GPU server · Nothing is stored
A short clip processed by the full pipeline: every person is detected and annotated frame by frame. (Friends intro, AI-upscaled, used here only as a sample.)