Descripción del proyecto
Interest in Virtual Humans is growing fast in public opinion and in several scientific and economic fields, from entertainment to medicine, from social sciences to ergonomics. The market of Virtual Avatars has a value estimated around USD 10 Billion and is expected to reach more than USD 520 Billion in 2030. To investigate, learn from, and represent fairly the wide variety of human experiences, tools to establish analogies or, namely, correspondence between them are required. Computer Vision and Graphics studied the problem intensely, but so far, no method has affirmed itself as a robust and flexible standard. This ambitious action aims to fill this gap, starting from three observations: 1) few methods rely on implicit representations, despite their flexibility and resilience; 2) especially, well-studied theoretical methods capable of strong regularizations have never been applied to these representations; 3) no method takes full advantage of large-scale datasets.
This action will combine these aspects, following a roadmap with three major scientific objectives:
a) Collecting dataset of 10 Million bodies with different properties, encoded as implicit representation, equipped with a ground-truth correspondence;
b) Developing a novel data-driven framework based on Functional Maps theory for implicit representations;
c) Deploying a large-scale neural network for 3D Human Correspondence beneficiary of the previous bullets, usable in real-world scenarios
This MSCA will produce substantial scientific, economic, and social impacts in this strategic field thanks to the interdisciplinary union of mathematical tools for functional analysis, the latest advances in deep learning, and domain knowledge of human bodies. This action will carry an intense outreach to a broad audience, informing on bodies digitalization process and the importance of fair representation of the human experience in this fast-changing technology.