First Workshop on Deep Learning Foundations
of Geometric Shape Modeling and Reconstruction
Seattle, USA | June 19th, 2020
Aim & Scope
Following the success of semantic and reconstructive image analysis, the landscape of geometric shape modeling has been largely changed with the recent surge of deep learning based data-driven solutions. New techniques specific to 3D shape learning keep emerging, from 3D CNNs and their efficient variants, to graph neural networks and the more recently ones that learn on implicit shape fields. Researchers are also applying these techniques to a variety of shape modeling tasks, ranging from shape reconstruction, restoration, shape conversion, to even surface rendering; these techniques are proposed to be compatible with different forms of shape approximations, with point set, volume, and mesh as the prominent representatives.
While most of these technical innovations are leveraging the great modeling capacities of deep networks and massive amounts of training shapes, however, they tend to overlook fundamental issues related to learning and modeling of geometries and shapes. These learning foundations may include notions or properties that enable surface modeling as continuous manifolds, algebraic or topological complexities of object surfaces, approximation error bounds of various shape representations, efficient and effective shape encoding, generalization bounds characterizing conditions of novel surface reconstruction, architectural designs and solution properties of deep networks that are optimal for learning geometries and shapes, and many others.
These learning foundations are of useful principles to a variety of current hot topics and problem settings in learning based shape reconstructions, e.g., from pure learning based ones to multi-view geometric ones. Bearing these foundations in mind would help integrate the relevant learning principles into algorithmic designs. This workshop aims to bring together researchers working in this frontier to discuss and address the aforementioned fundamental issues. The workshop also aims to bring attention of the community on these geometric learning foundations. Topics include, but not limited to:
● Geometric shape modeling foundations
● Theoretical analysis of surface properties
● Topological analysis of surfaces and shapes
● Surface complexity analysis
● Shape representations and their approximation errors
● Studies on shape encoding
● Generalization analysis on learning novel surfaces
● Optimization analysis of graph networks