I grew up in the sunny city of
During my Ph.D. in physics-based simulation with Nils Thuerey, we used the convex optimization technique fast first-order Primal-Dual method [PCBC09] for several complex problems, such as reconstructing both 3D volume and motion of real-world fluid phenomena based on 2D input sequences, which is the inverse problem to forward fluid simulation. We targeted accurate multi-view reconstructions from a sparse number of cameras, which we gather in our data set ScalarFlow, as well as single-view reconstructions, for which we won the best paper award from SCA'18. Furthermore, we developed a flexible fluid guiding method and achieved separating boundary conditions for liquids with a common CG solver. Have a read through my Ph.D. thesis!
I organized the exercises for the lectures "Basics: Algorithms and Data Structures" and "Deep Learning and Numerical Simulations for Visual Effects" as well as the seminar "Deep Learning in Physics". Additionally, I supervised Bachelor's and Master's Theses. Keywords: fluid simulation, fluid capture, convex optimization, numeric solvers, neural networks
Currently, I am a Simulation Researcher at Weta Digital working on incredibly interesting and fascinating challenges. We went to SIGGRAPH Asia '19 togther!
Besides working on numerical solvers for physics-based simulation, I love
In my Ph.D. Thesis "Optimization for Fluid Simulation and Reconstruction of Real-World Flow Phenomena" [Eck19], I target the improvement of both realism and control of fluid simulations by using convex optimization with the fast first-order Primal-Dual method [PCBC09]. The following three papers were developed in the scope of my Ph.D.: accurate reconstruction of real-world fluid phenomena, the highly under-determined single-view reconstructions of real-world smoke plumes, and efficient fluid guiding and separating solid-wall boundary conditions.
ScalarFlow [EUT19] is the first large-scale data set of accurately reconstructed real-world smoke plumes. We collected more than 100 3D reconstructions in our Data Set.
Besides the data, we present a framework for accurate physics-based reconstructions from a small number of video streams. Central components are a novel estimation of unseen inflow regions and an efficient optimization scheme constrained by a forward simulation.
In order to more flexibly use different recording devices and already available data sets, we propose a
Single-View Reconstruction Method [EHT18] for real-world smoke phenomena. Without any information in the depth direction, the reconstruction problem is heavily under-determined. However, using powerful physical priors and inferring density updates strongly coupled to previous and current estimates of the flow motion, our method produces realistic fluid motions.
We introduce the fast first-order Primal-Dual method [PCBC09] to the fluids community by proposing a solution for two complex problems:
Fluid Guiding and Separating Solid-Wall Boundary Conditions [IEGT17] for liquids. Convex optimization allows us to split complex problems into simpler problems, which are usually easier to solve. Hence, we achieve explicit control over both large-scale motions and small-scale details, and effectively eliminate unrealistic artefacts of fluid crawling up solid walls and sticking to ceilings, while requiring only few changes to existing implementations.
In this Master's Thesis [Eck14], I develop an efficient approach of solving for separating solid-wall boundary conditions for liquids. Allowing for positive normal velocity components at fluid-solid faces forms an inequality constraint and usually transforms the linear equation system for the pressure solve into a Linear Complementarity Problem, which is typically solved by expensive Quadratic Pro-gramming solvers. However, a more efficient approach can be realized by using convex optimization with the alternating direction method of multipliers (ADMM).
During my student's project at Eurecom with Prof. Dr. Jean-Luc Dugelay, we investigated the Impact of Facial Cosmetics on Automatic Face Regognition [EKD13]. Our studies show that contrast-enhancing makeup can increase the accuracy of face recognition, while heavily perceived-shape altering impairs the accuracy, as expected. We provide a data set of several individuals with varying amount and types of makeup.
For my Bachelor's Thesis, we investigated Different Strategies for Application Landscape Integration in Mergers and Acquisitions [EFMRS12]. When two companies unite, the essential question of how to unify the different application landscapes arises. The successful integration is crucial for both maintaining service efficiency but also for the motivation of employees. We interviewed multiple industry experts and developed a metric to compare the most common strategies.
SIGGRAPH Asia, Brisbane, November 2019
Technical-papers presentation of "ScalarFlow: A Large-Scale Volumetric Data Set of Real-world Scalar Transport Flows for Computer Animation and Machine Learning" [EUT19]
SCA, Paris, July 2018
Technical-papers presentation of "Coupled Fluid Density and Motion from Single Views" [EHT18] - best paper award
Albert-Ludwigs-University Freiburg, May 2017
Invited Talk with Prof. Dr.-Ing. Matthias Teschner about "3D Reconstruction of Volume and Velocity of Real Fluid Phenomena Based on a Single Camera View"
KAUST, April 2017
Technical poster presentation of "Reconstructing Volume and Motion from Real Fluid Phenomena with a Minimal Number of Camera Views", invited by Prof. Dr.-Ing. Wolfgang Heidrich
Eurographics, Lyon, April 2017
Technical-papers presentation of "Primal-Dual Optimization for Fluids" [IEGT17]
MMSP, Sardinia, September 2013
Technical poster presentation of "Facial Cosmetics Database and Impact Analysis on Automatic Face Recognition" [EKD13]
SIGGRAPH, SIGGRAPH Asia, Computer & Graphics Journal (CAG)
[Eck19] ECKERT M.-L.: Optimization for Fluid Simulation and Reconstruction of Real-World Flow Phenomena. PhD Thesis (2019).
[EUT19] ECKERT M.-L., UM K., THÜREY N.: ScalarFlow: A Large-Scale Volumetric Data Set of Real-world Scalar Transport Flows for Computer Animation and Machine Learning. ACM Transactions on Graphics (2019).
[EHT18] ECKERT M.-L., HEIDRICH W., THÜREY N.: Coupled Fluid Density and Motion from Single Views. Computer Graphics Forum 37 (2018).
[IEGT17] INGLIS T.*, ECKERT M.-L.*, GREGSON J., THÜREY N.: Primal-Dual Optimization for Fluids. Computer Graphics Forum 36 (2017).
[Eck14] ECKERT M.-L.: Flexible Boundary Conditions in Fluid Solvers Based on Proximal Operators. Master's Thesis (2014).
[EKD13] ECKERT M.-L., KOSE N., DUGELAY J.-L.: Facial Cosmetics Database and Impact Analysis on Automatic Face Recognition. Multimedia Signal Processing (2013).
[EFMRS12] ECKERT M.-L., FREITAG A., MATTHES F., ROTH S., SCHULZ C.: Decision Support for Selecting an Application Landscape Integration Strategy in Mergers and Acquisitions. European Conference on Information Systems (2012).
[PCBC09] POCK T., CREMERS D., BISCHOF H., CHAMBOLLE A.: An Algorithm for Minimizing the Mumford-Shah Functional. International Conference on Computer Vision (2009).
Winter 2018: Seminar Deep Learning in Physics
Summer 2018: Advanced Deep Learning for Physics
Summer 2019: Basics: Algorithms and Data Structures
Summer 2017: Deep Learning and Numerical Simulations for Visual Effects
Summer 2016: Simulation for Visual Effects
Summer 2015: Simulation for Visual Effects
“Learning to Reconstruct Smoke Volumes from Images” – Daniel Frejek, Master’s Thesis, 2019
“Improving a Low-Cost Capturing Process for Reconstructing Volume and Motion of Real Fluid Phenomena” – Daniel Frejek, Guided Research, 2017
“Capturing Real Fluid Phenomena with Raspberry Pi Cameras” – Florian Alkofer, Bachelor’s Thesis, 2017
“GPU-accelerated Stochastic Tomography for 3D Volume Reconstruction of Real Fluid Phenomena” – Tobias Kammerer, Master’s Thesis, 2017
“Optimized Volume Reconstruction for Fluids with Non-Linear Lighting Models” – Tobias Gottwald, Master’s Thesis, 2017
“Experimental Capture of Smoke and Evaluation of Volume Reconstruction Algorithms” – Florian Reichhold, Master’s Thesis, 2016
“Reconstruction of Fluid Volumes Based on Stochastic Tomography” – Dominik Dechamps, Master’s Thesis, 2016
“Modeling 3D Fluid Volumes Based on Appearance Transfer” – Christoph Pölt, Master’s Thesis, 2015