Figure 6

Camera position estimates in manifold lifting demonstration. a Initial ISOMAP embedding v₁, v₂, . . ., v _J of the measurement vectors. b Initial estimates $\left\{\widehat{{\theta }_j}\right\}$ of camera positions after rotating and scaling the {v _j }. c Final camera position estimates after running the manifold lifting algorithm. In b and c, the colored points represent the estimated camera positions (color coded by the true ${\theta }_j^H$ value), while the blue vectors represent the error with respect to the true (but unknown) camera position.