YOLO's layers output

YOLO's final output is a w × h × M matrix, where w × h is the size of the grid, and M corresponds to the formula B × (C + 5), where the following applies:

• B is the number of bounding boxes per grid cell.

• C is the number of classes (in our example, we will use 20 classes).

Notice that we add 5 to the number of classes. This is because, for each bounding box, we need to predict (C + 5) numbers:

• t_x and t_y will be used to compute the coordinates of the center of the bounding box.

• t_w and t_h will be used to compute the width and height of the bounding box.

• c is the confidence that an object is in the bounding box.

• p1, p2, ..., and pC are the probability that the bounding box contains an object of class 1, 2, ..., C (where C = 20 in our example).

This diagram summarizes how the output matrix appears:

Figure 5.5: Final matrix output of YOLO. In this example, B = 5, C = 20, w = 13, and h = 13. The size is 13 ×...