It seems to do most of its work by making assumptions based on the user input, so it's better for man made objects I would think. I remember another project that did a similar thing for interior spaces, in what they called a "Manhattan World". You can assume that doors, windows, walls, etc will be straight lines and right angles and fill in a lot of gaps from there.I would be handy for getting distances and dimensions of things from a single photo. If you can give it one reference distance, it could tell you a lot about the sizes of a lot of other things in the frame.
]]>