Skip to content

Image Classifier

Train an image classifier in your browser — no code, no Python, no cloud GPU needed.

Features

  • Class management — add named classes, drag images into each
  • Transfer learning — uses MobileNet as feature extractor, trains a custom head
  • Live training — see loss/accuracy during training
  • Test predictions — drop an image or use webcam, see confidence per class
  • Export — download trained model as TF.js format

How it works

  1. Create 2+ classes (e.g. "cat", "dog")
  2. Add 5+ images per class (drag-and-drop or file picker)
  3. Click Train — MobileNet extracts features, dense head trains on your classes
  4. Test with new images or webcam
  5. Export the model

Technical details

  • MobileNet v2 as frozen feature extractor (14 MB download, cached)
  • All images resized to 224x224
  • Dense classifier trained on MobileNet embeddings
  • Typical training: 10-30 seconds for 50-200 images