Mastering Computer Vision with TensorFlow 2.x

Region proposal

In this stage, the algorithm uses the selective search method described previously to extract around 2,000 category-independent region proposals. The category-independent region proposals are used to identify a number of regions in an image so that each object is well represented by at least one region in the image. Humans do this naturally by localizing objects in an image, but for machines, the location of the object needs to be determined and then it needs to be matched with an appropriate region in an image to detect the object.

Unlike image classification, detection involves image localization so that an appropriate region can be created that encloses the object to detect features within the region. The appropriate region ...