◉ YOLOE Explorer A LOOK INSIDE COMPUTER VISION

What is YOLOE?

YOLOE is an AI model that finds objects in pictures. Give it a picture and tell it what to look for, such as a cup or a chess piece. It draws boxes around possible matches and gives each one a score. You can change what it looks for by changing your words.

Celeste uses YOLOE to search its camera pictures. In this app, you can try a saved camera picture or upload your own, then see how the model uses the picture and your words to guess where the object is. The predictions can be wrong, so compare them with what you see.

How Celeste connects a camera picture with your words The picture goes through the backbone and neck. Your words go through the text encoder. They meet at the detection head, which produces boxes, labels, and scores. Hover, focus, or tap a component to learn what it does. Current camera picture Loading picture Backbone: build patterns from the picture.Builds patterns from the picture, starting with simple details and combining them into more complex patterns. Changing your words does not change these picture patterns. Backbonebuild picture patterns Neck: bring details and bigger shapes together.Brings together small details and bigger shapes, helping the detector find objects that appear at different sizes in the picture. Neckcombine detail + context Detection head: connect your clue with places in the picture.One part uses picture patterns to suggest box edges. Another compares picture patterns with your words to score possible matches. Your words change the scores, rather than moving the suggested edges. Detection headmatch words to placessuggest box edges Predictions: the model's guesses.The model's guesses: boxes show where, labels show which clue matched, and scores show how strong the matches are. Check these guesses against what you can see. Predictionsboxes + labels + scores cup
cup
 Text encoder: turn your words into a number pattern.Turns your words into numbers and prepares them for comparison with picture patterns. It does not choose the box edges. Text encoderturn words into a number pattern
Parts of YOLOE

Select a component to see what it does.