"Stupid vision tricks". Or you need a hook to perk up interest in your paper for citation .
The real meat was in proving that you didn't need highly controlled rigs and calibration to snap images for the Surface from Motion software/algorithms they already had. That and that you could generate 3D point clouds from really lo-res image sets.
Practical applications would be in automation E.G. robot navigation and kinematics. The paper mentioned it was possible to do SFM on the phone, but it would have snapped up so many resources as to render the phone incapable of doing any other task.
ADDENDUM: The other practical application is using exisiting information which is overwhelming "lots 'O images" into an intuitive point-n-click interface. E.G. Big Data graphical frontend.
Keep Up with TechRepublic