We propose an algorithm for generating an unstructured lumigraph in real-time from an image stream. This problem has important applications in mixed reality, such as telepresence, interior design or as-built documentation. Unlike conventional texture optimization in structure from motion, our method must choose views from the input stream in a strictly incremental manner, since only a small number of views can be stored or transmitted. This requires formulating an online variant of the well-known view-planning problem, which must take into account what parts of the scene have already been seen and how the lumigraph sample distribution could improve in the future. We address this highly unconstrained problem by regularizing the scene structure using a regular grid structure. Upon the grid structure, we define a coverage metric describing how well the lumigraph samples cover the grid in terms of spatial and angular resolution, and we greedily keep incoming views if they improve the coverage. We evaluate the performance of our algorithm quantitatively and qualitatively on a variety of synthetic and real scenes, and demonstrate visually appealing results obtained at real-time frame rates (in the range of 3Hz-100Hz per incoming image, depending on configuration).

In this paper, we show that genetic algorithms (GA) can be used to control the output of procedural modeling algorithms. We propose an efficient way to encode the choices that have to be made during a procedural generation as a hierarchical genome representation. In combination with mutation and reproduction operations specifically designed for controlled procedural modeling, our GA can evolve a population of individual models close to any high-level goal. Possible scenarios include a volume that should be filled by a procedurally grown tree or a painted silhouette that should be followed by the skyline of a procedurally generated city. These goals are easy to set up for an artist compared to the tens of thousands of variables that describe the generated model and are chosen by the GA. Previous approaches for controlled procedural modeling either use Reversible Jump Markov Chain Monte Carlo (RJMCMC) or Stochastically-Ordered Sequential Monte Carlo (SOSMC) as workhorse for the optimization. While RJMCMC converges slowly, requiring multiple hours for the optimization of larger models, it produces high quality models. SOSMC shows faster convergence under tight time constraints for many models, but can get stuck due to choices made in the early stages of optimization. Our GA shows faster convergence than SOSMC and generates better models than RJMCMC in the long run.