Cat detected · 98%: how AI checks every photo

Under the hoodSeptember 25, 2026

CatOnMap galleries have no random selfies, food shots or screenshots — only cats. That's not magic but a pipeline of neural networks checking every photo before publication.

What happens after you press the button

When you take a shot, it flies to the server and enters the AI review queue. A detector model looks for cats and people in the frame, while a separate model checks for inappropriate content. All of it takes a fraction of a second — faster than you can open the gallery.

The rules are simple

  • ✅ A cat is present, no people, clean content — the photo is published automatically. No waiting for a moderator.
  • ❌ No cat found — the shot is rejected. Sometimes it's a close call (more on that below).
  • ❌ A person in the frame — even if there's a cat too. We deliberately don't publish people: it's both privacy and gallery hygiene.

Night cats and distant cats

The hardest case for the detector is a cat at night, in the shadows or in a far corner of the frame. We keep tuning the pipeline on such cases: recently we lowered the detector's confidence threshold and switched analysis to full resolution — and the "night" cats the system used to discard started passing the check.

What if the AI got it wrong?

Borderline shots go to human moderation — a real person reviews them. Until the decision, only you can see the photo; once approved, it joins the country gallery and earns a point in the current battle.

So when the detector says "Cat detected · 98%", you can be sure: there really is a cat in your collection.

← All posts