The paper says: """ We compute a mask for the object pixels, and use this mask to inpaint the background using the PatchMatch algorithm [Barnes et al. 2009]. F

The paper says: """ We compute a mask for the object pixels, and use this mask to inpaint the background using the PatchMatch algorithm [Barnes et al. 2009]. For complex backgrounds, the user may touch up the background image after inpainting. """ So they're not really claiming perfect object removal is part of their thing. What they are actually doing is quite interesting, but their pitch is making some rather extravagant claims IMO.