A new technique can model an entire 3D scene, including areas hidden from view, from just one camera image. The method, developed by MIT and Meta researchers, relies on image shadows, which provide information about the geometry and location of hidden objects.