Top pic is an image I use. Originally, size of that pic is 100 by 133. I use photoshop to change campus size to 640 by 480. ( try not stretch image itself ).
Middle one is I did not train AAM. I just get parameters after I updated landmarks.
Bottom one is I re-trained AAM, and got parameters after re-training.
I used AAM search to see if it could find Obama face, but it failed in both cases. ( re-train or not ).