Image AI · Transfer learning
Cat or dog?
Show it a photo and a neural network decides. It runs entirely on your device, so your photos never leave it.
Drop a photo here
or use the buttons below. You can also paste an image.
Prediction
Loading the model…
—
—
How it works
- Resize. Your photo is shrunk to 160 × 160 pixels: 76,800 numbers.
- See. EfficientNet-B0, a network pretrained on over a million ImageNet photos, turns it into 1,280 numbers describing what it sees: fur, ears, eyes, face shape.
- Decide. A single trained neuron weighs those 1,280 numbers and outputs the probability of "dog".
- Answer. Above 50% means dog; below means cat.
About the model
- Architecture
- EfficientNet-B0 (frozen) + 1 sigmoid neuron
- Trained on
- 18,610 photos
- Test accuracy
- 98.80% on 2,326 unseen photos
- ROC AUC
- 0.9986
- Download size
- ≈ 16 MB, once
- Runs on
- Your device, with TensorFlow.js
Good to know
- It only knows two answers. Show it a person or a car and it will still pick cat or dog.
- Works best when one animal is clearly visible and fills much of the photo.
- Hard cases: tiny animals far away, very dark cats, fluffy small dogs, and photos with both a cat and a dog.