Virtualization of Real Flows for Animation and Simulation

Projektinformationen

realFlow

ID Finanzhilfevereinbarung: 637014

DOI

10.3030/637014

Projekt abgeschlossen

EK-Unterschriftsdatum 9 Februar 2015

Startdatum 1 Mai 2015

Enddatum 31 August 2020

Finanziert unter

• EXCELLENT SCIENCE - European Research Council (ERC)

Gesamtkosten

€ 1 465 603,75

EU-Beitrag

€ 1 465 603,75

Koordiniert durch

TECHNISCHE UNIVERSITAET MUENCHEN
Germany

Periodic Reporting for period 4 - realFlow (Virtualization of Real Flows for Animation and Simulation)

Berichtszeitraum: 2019-11-01 bis 2020-08-31

Zusammenfassung vom Kontext und den Gesamtzielen des Projekts

This ERC StG project is titled “realFlow - Virtualization of Real Flows for Animation and Simulation". It's goal is to improve the simulation of physical processes and, above all, make it possible to generate such simulations more quickly. The focus here is on liquids and gases, so-called fluids.

Simulations of fluids can be very useful in many applications – for example in the design of aircraft wings and the depiction of explosions in films and computer games. Blood flow simulations in medicine can help in the assessment of the threat posed by aneurysms. We plan to create a database containing previously-calculated simulations and video recordings of the behavior of real fluids. This information would then be available for new simulations. The project involves, firstly, the development of machine learning algorithms that will enable the processing, comparison and classification of enormous volumes of data. Video recordings of the actual liquids or gases must also be analyzed and transformed into compatible data. Ideally, this will lead to interactive and stable simulations of complex physics effects.

Overall, this project has been very successful: we have significantly advanced the state of the art in terms of algorithms for capturing and digitizing fluids and for data-driven methods. Among others, a first public large-scale data-set of flow data was published within the scope of this project (the ScalarFlow dataset).

Arbeit, die ab Beginn des Projekts bis zum Ende des durch den Bericht erfassten Berichtszeitraums geleistet wurde, und die wichtigsten bis dahin erzielten Ergebnisse

We have developed methods for flow reconstruction using coupled optical-flow and physics simulations. We could show that this combination significantly out-performs purely visual optical flow reconstructions, which illustrates one of the central themes of this research project: the usefulness of physics simulations to disambiguate measurements. In addition, we have developed a fluid simulation methods to perform simulations with pre-computed space-time data sets. Thus, instead of re-running simulations from the start, this makes first steps towards the “recycling” of simulation data sets. Additional projects have targeted controllable and example-based simulations, which work hand-in-hand towards the goal to establish data-driven methods for flow simulations, and other physical phenomena.

Fortschritte, die über den aktuellen Stand der Technik hinausgehen und voraussichtliche potenzielle Auswirkungen (einschließlich der bis dato erzielten sozioökonomischen Auswirkungen und weiter gefassten gesellschaftlichen Auswirkungen des Projekts)

Progress beyond the existing state of the art was made in several areas:

- making it possible to employ so called “stream functions” to simulate liquids,

- controlling the plastic deformation of elastic objects with rigged examples,

- enabling correct dispersion effects in real-time water wave simulations,

- extracting the physical parameters of collisions from video,

- and robustly interpolating whole space-time data sets of liquid and smoke simulations.

Despite impact in the area of computer graphics research, the methods developed within the scope of realFlow have already had significant interdisciplinary impact in fields of computer graphics, vision, machine learning and computational science.

A temporally coherent high resolution volume generated by a deep neural network (l: in, r: out)

Different variations of a fluid simulation guided with our convex optimization approach.