New algorithms facilitate complex video processing such as automatic dynamic background inpainting. New machine learning algorithms make possible the removal of compression artifacts from images and videos.
With the work of Anna and Markus Rohrbach the automatic generation of subtitles or the automatic movie description for the blind becomes a realistic objective.
The 3D reconstruction of detailed models of persons in general apparel is, until now, possible in indoor sudios with controlled environment only. With new algorithms this seems it could succeed even in arbitrarily complex outdoor scenes with only a few camera.
Instead of using a large variety of classifier types, it is possible to obtain best results in shortest time by precisely applying the sliding windows technique.
Using a new design model (3D object class detector) not only the efficient detection of objects and their 2D position in a picture is possible, but also their relative 3D position and view point estimation.
We developed a novel mathematical approach that allows streak and time lines to be described using ordinary differential equations therewith overcoming the drawbacks of former complex algorithms.
Olga Kalinina combines genomic-, modeling-, and biophysics-based methods to assess the mutations observed in given sequences of viral proteins and infer their potential influence on the function of the protein. This enables suggestions for drug treatment of viral diseases.
We could improve the bounds of numerical steps needed to find an exact solution for polynomial systems by multiple orders of magnitude. Given present knowledge, they are nearly optimal.
The interrelation between supply and demand is easily described as long as you have just one good. For the complicated calculations in an economy with many goods, many suppliers, and many buyers Ran Duan and Kurt Mehlhorn found a relatively simple combinatorial algorithm.
Processing "Big Data" poses new challenges to computer science. At the Max Plack Institute for Informatics, we are investigating algorithms and systems to mine useful information from large, complex data sources efficiently and effectively.
Capturing motion pictures and representing them as 3d objects could until now only be efficiently performed by utilizing marker-based camera systems. Using new algorithms developed by Professor Theobalt and co-workers actors are no longer constricted by marker suits.
