Image Classification

See the [[Computer Vision Week 1 Main File]] or the [[Main AI Page]]

Like a human describing an image, Computer Vision Image Classification takes an image and lists objects/identifiers the used model can perceive in the image, labelling them, and usually along with a percentage representing how certain it is that the image contains those labels.

Confidence Scores

These percentages are called confidence scores and represent how confident that the image contains the label. They are expressed as two-decimal numbers in the range of 0.00 to 1.00.

These scores are generated using a sigmoid normalisation function, so 1.00 is more than two times the confidence of 0.50

Breadth Vs Depth

A classifier that is good at classifying a wide range of general classifiers, is said to have more breadth than depth.

A classifier that is good at classifying specific domains of images with a great deal of specificy has more depth than breadth.