The modelling of the visual content of natural and artificial objects is of paramount importance in many fields of science and engineering.
In particular, the automatic inference of the geometry of shapes given solely images as an input has brought a tremendous impact in several application fields in science. In this regard, PAVIS brings its expertise in a pervasive manner through the research of advanced Computer Vision and Signal Processing methods. To cite a few examples, starting from the tiniest elements, we collaborate with the IIT Nanophysics department to push forward the boundaries of 3D super-resolution in microscopy. Together with the NBT department, we are researching methods for modelling the geometry of in-vitro neuronal networks in order to pursue a more detailed understanding of their function. At a larger scale, we are developing algorithms to geometrically self-localise sensors (microphones, cameras) freely deployed in an area and linked together by a network. We are also actively studying severely ill-posed problem such as the geometrical modelling of non-rigid shapes solely from 2D images. We have also developed efficient methods for accurately estimating the 3D shape of larger scale objects (up to an asteroid in size) using Structure from Motion approaches. All these different applications shares a common aspect which is the underlying geometry sensed by heterogeneous devices. Our aim is to provide effective methodological approaches to obtain the most accurate and complete inference of the 3D geometry of the world from the tiniest elements to the most macroscopic shapes.