Machine Learning for Developers

Detection

Detection adds a level of complexity because it requires guessing the location of one or more elements pertaining to the image, and then trying to classify each of these elements of information.

For this task, a common strategy for the individual localization problem is to combine a classification and regression problem—one (classification) for the class of the object, and the remaining one (regression) for determining the coordinates of the detected object—and then combining the losses into a common one.

For multiple elements, the first step is to determine a number of regions of interest, searching for places in the image that are statistically showing blobs of information belonging to the same object and then only applying the ...