cracksmity.dev
Español
L2 · Friend Station In service

Cat Detector

To wrap a gift in photos of her cat, I had to find them among 12,000 images in a chat. So I put CLIP to work for me.

OPENED
JUL 2026
FOR
A friend
SOURCE
Public on GitHub

Code on GitHub LinkedIn post

Light window of the cat image filter: source and destination folders, a confidence threshold at 80%, test mode and classification progress.
The detector: source folder, destination folder and how sure it has to be that there's a cat.

ROUTE · L2 Friend · 3 stations

  1. JUN 2026 · In service MC-Barra-Salud
  2. YOU ARE HERE Cat Detector
  3. SEP 2026 · In service Sticker Studio

Why it exists

I always give my friend her presents in creative wrapping, and this year she gave me the idea herself: use photos of my cat. The problem was they were buried in a WhatsApp group among more than 12,000 images, memes and screenshots. Instead of 15 minutes of scrolling, I spent hours building an app.

It walks a folder, classifies every image with CLIP zero-shot and copies only the cat pictures. It got about 80% right, wrote a CSV report and ran on my AMD card through DirectML, after a fight with Python versions. Some false positives were definitely not cats, but I found every photo of her cat and the gift got wrapped.

How it's built

  • CLIP
  • PyTorch
  • DirectML
  • Python