Learning to classify
An image classifier is trained with labelled examples: many photos marked “ripe” and many marked “unripe”. During training it adjusts itself to tell them apart.
Early layers of a vision model respond to simple things such as edges and colours. Later layers combine these into textures, shapes and eventually whole objects.
When you show it a new image, it outputs a label with a confidence score, for example “ripe, 87%”. That score is the model’s estimate, not a guarantee.
A classifier returns “cat, 62%”. What does 62% mean?
Choose the answer you think is right.