Onur Tasar (TITANE)
POLYGONIZATION OF BINARY BUILDING CLASSIFICATION MAPS USING MESH APPROXIMATION WITH RIGHT ANGLE REGULARITY
One of the most popular and challenging tasks in remote sensing applications is the generation of digitized representations of Earth’s objects from satellite images. A common approach to tackle this challenge is a two-step method that first involves performing a pixel-wise classification of the satellite images, then vectorizing the obtained classification. In this talk, I will explain the proposed approach to vectorize input binary building classification map by optimizing a labeled triangle mesh, which minimizes an objective function that balances between fidelity to the classification map in L1 norm sense, right angle regularity for polygonized buildings, and final mesh complexity.
Johanna Delanoy (GRAPHDECO)
How to make computers understand human sketches? An introduction to sketch based modeling
Modeling an object in 3D can be a tedious process since it requires a good knowledge of the software and of the modeling process. On the other hand, sketching is a very natural way to represent objects. The ambition of sketch-based modeling is to bring the ease and immediacy of sketches to the 3D world. However, while humans are extremely good at perceiving 3D objects from line drawings, this task remains very challenging for computers. In this talk, I will present various strategies that have been developed to simplify this problem in order to build interactive sketch based modeling systems. I will then present our new approach which tries to overcome existing limitations by using deep learning at the core of our system.