What does this do?
The program first calculates the Structure-From-Motion for the sequence of uploaded images. This includes a self-calibration of the camera's internal properties such as the focal length.
If this first stage is completed successfully then a dense 3D reconstruction (using a novel GPU-accelerated algorithm) is made of the scene. The output displayed in the '3D Interpretation' viewer is of a very low-quality and textured from just one of the images (to aid download speeds and portability). However, this is generated from a much higher quality textured polygon model. More high-quality outputs will be available in the short-term.