Performing non-max suppression

So far, in the previous section, we have only considered the candidates that do not have a background, and further considered the candidate that has the highest probability of the object of interest. However, this fails in the scenario where there are multiple objects present in an image.

In this section, we will discuss the ways to shortlist the candidate region proposals so that we are in a position to extract as many objects as possible within the image.

Getting ready

The strategy we adopt to perform NMS is as follows:

• Extract the region proposals from an image
• Reshape the region proposals and predict the object that is contained in the image
• If the object is non-background, we shall keep...