Non-Max Suppression

A common problem of object detection algorithm is that they tend to detect an object multiple times even if it is only present once in the image.

Non-Max Suppression is a way to fix this.

Consider the following image with multiple bounding boxes generated per object:

As shown, each bounding box is associated with a probability. Non-Max Suppression is the process of eliminating the bounding boxes that have non-max probability for each object, thereby only retaining one bounding box per object.