Instead of working from a single 2D image as this work does, what is the state of the art for 3D reconstruction from multiple (non-calibrated) images? e.g. some casual shots of an object using my phone.