Before, I used 2D gaussian fitting as aperture photometry, but that is wrong.
I put too many constrain on the model.
Now I am working on construct a correct one.

I purpose to collect a region, and just count the flux from, don't apply any model.