Done by training an initial model on a dataset, having the technology detect semantic keywords automatically, and then using human effort to correct its output and then retrain the model.

Image Classification is an annotation style that will label or classify a whole image or scene, so it best works when there is a single object being defined.

Bounding boxes are the most commonly used type of annotation. They are rectangular boxes which are defined by x and y axis coordinates of the upper left hand corner and bottom right hand corner. This type of annotation indicates the spatial location of the object relative to the background.

Polygonal segmentation annotations are complex polygons that are used to be more accurate than bounding boxes and indicate the shape of the object.