Attention area when CNN

Now there are:

  • Attention map (the 2D matrix of important value)
  • Associated original image

Then I want to implement CNN, which pays more attention to the location of original image on the basis of attention map.

How can I achieve this idea?