Putting the User in the Loop for Image-Based Modeling

摘要: We refer to the task of recovering 3D structure an object or a scene using 2D images as image-based modeling. In this paper, we formulate discrete optimization problem solved via energy minimization. standard framework Markov random field (MRF) defined over image present algorithms that allow user intuitively interact with algorithm. introduce algorithm where guides process modeling find and model interest by manually interacting nodes graph. develop end applications on mobile device printing interest. also propose alternate active learning input. An initial attempt is made at reconstructing without supervision. Given reconstruction, uses intuitive cues quantify uncertainty suggest regions, querying provide support for uncertain regions simple scribbles. These constraints are used update unary pairwise energies that, when solved, lead better reconstructions. show through machine experiments study proposed approach intelligently queries users constraints, achieve reconstructions faster, especially scenes textureless surfaces lacking strong textural structural typically require.