DinoDreamer

How it works

DINO is a vision transformer that learned to see without labels. It turns a picture into 256 descriptions, one for each patch. A basis collection reads each description with 16,384 small learned spaces, three numbers wide: the basis groups. A cellular automaton is then trained until DINOv3 reads the image it grows as a chosen patch. This page follows one picture of a hibiscus through all three. After it, open the Index, watch the Automata, or Train one. The longer account is the write-up, Growing cellular automata to visualize DINOv3.