Figure 3: The computation of pixels in attention regions.