CNN for Object Detection

When we previously discussed image classification, we were interested in assigning a single category to an image, assuming it showed a single foreground object belonging to one of the classes in a predefined list. Image classification doesn’t supply any information about the object’s location in the image.

Object detection is the task of locating the objects in an image. It could be used, for example, to locate the chips in the image of an electronic board.

Object detection supplies a bounding box per object. We’ll get the object’s rough location but not the object’s contour. If the task requires obtaining the object contour, such as computing the object orientation in the image plane, we should perform semantic segmentation instead. ...