This PhD topic is part of the SIMILAR-CITIES research project funded by the French Funding: National Agency for Research (ANR) / project REVES

This project proposes to study the problem of texture management as a whole, in the context of urban scenery. Our key insight is that in many cases only a small subset of the scene must be strictly identical to the reality. For instance, landmarks and historical buildings must be accurately reproduced in all details. However, for the vast majority of the other parts of the scene we only seek for a plausible reproduction, not a strictly similar one. Instead of reducing the resolution of the images on these buildings, hence introducing a very noticeable lack of details, we propose to capture the appearance of the buildings with a procedural model. That is, a compact data structure encoding only the important information (positions of windows, doors, type of materials) without capturing the details (the position of each brick, each cracks). The details will be generated only at the time of display, hence not requiring any storage. We can see that our approach has to consider the problem of content management as a whole: We need to build procedural models from photographs, store them in compact data structures, load this data progressively as the user explores the scene, and finally display it at interactive frame rates. Each step corresponds to scientific and technological challenges: Capturing important information in a photograph (i.e. separating semantic from detail), studying data structures allowing compact storage and fast decoding, determining what will remain hidden or what will become visible for efficient loading, and finally generating details only in last resort to maintain small memory usage while enforcing interactive frame rates.

More details here

French version here 

Contact: Sylvain Lefebvre

Loop computation in [1] From left to right: Input model, handle (green) and tunnel (red) loops, handle features (green) ( Image copyright: Dey et.al)

Master level project proposal: optimal loop computation on surfaces

Contact: This e-mail address is being protected from spambots. You need JavaScript enabled to view it