Generating realistic structure-from-motion datasets through virtual photography

The paper presents the virtual photography method of generating structure-from-motion datasets. Virtual photography is an interactive process of making shots from different viewpoints and with arbitrary camera intrinsics according to the pinhole camera model. The result of virtual photography may be considered as input data for the structure-from-motion software. The paper provides realistic datasets according to the most significant properties of real datasets. Implementation of the virtual photography method using the OpenCV library is described.