Image recognition bench

Load a photograph and inspect the five strongest ImageNet labels from a compact local model.

How this experiment works

MobileNetV4 convolutions map a resized image to a learned distribution over 1,000 classes.

What it does not prove

The fixed ImageNet vocabulary does not cover every object. A high score can still be wrong.

Try this challenge

Try an unusual camera angle or crop. Which labels become more or less likely?

Related Trust Agent courses · Sources and licences

Enable JavaScript to run the experiment on your device.