Markus Plack

Researching high-resolution capture, dynamic 3D reconstruction, view synthesis, and real-time visual computing systems, with a strong focus on turning research ideas into robust, high-performance prototypes. • Architected and implemented distributed capture software for synchronized acquisition and live streaming of 24 MP visual data at 60 FPS. • Led software development for the Visual Computing Incubator, defining the architecture, coordinating work across 12 contributors, mentoring developers, and contrib

Conducted research in 3D computer vision and reconstruction, with a focus on spatial priors, stereo matching, non-line-of-sight imaging, and high-performance GPU implementations. • Developed GPU-accelerated differentiable transient rendering for non-line-of-sight reconstruction, enabling minute-scale optimization and resulting in a WACV 2023 publication. • Designed high-resolution stereo matching for volumetric capture, combining visual-hull priors with memory-efficient GPU processing; published at WACV 202

Developed a novel frame interpolation approach for rendered content using transformer-based modeling and uncertainty guidance. • Improved interpolation quality and robustness through uncertainty-aware motion estimation and synthesis. • Resulted in the CVPR 2023 publication "Frame Interpolation Transformer and Uncertainty Guidance".

• Improved optical flow estimation for rendered-content interpolation using a refined hierarchical flow-estimation network. • Increased robustness and precision of the underlying motion estimation approach. • Contributed to work later included in a Disney Research patent on frame interpolation for rendered content.

• Calibrated depth-camera systems for experimental computer vision setups. • Performed measurements to evaluate depth accuracy and sensor performance.

Researched topic modeling of short, dynamic text data as part of UCLA’s Applied and Computational Mathematics REU program and subsequent master’s thesis work. • Improved NMF-based topic models using distributed word representations. • Extended the approach to model temporal dynamics in short-text data. • Evaluated the method against baseline models across multiple NLP datasets. • Built an interactive visualization system for exploring topics across time and geographic location.

• Analyzed Raman microscopy data using spectral unmixing methods. • Evaluated the stability of unmixing algorithms on real and synthetic datasets. • Developed a visualization system to support estimation of the optimal number of spectral endmembers.

Doctoral thesis: Spatial Priors and Uncertainty for Enhanced Reconstruction in Computer Graphics and Vision Research focused on 3D reconstruction, stereo matching, non-line-of-sight imaging, GPU-accelerated differentiable rendering, and high-resolution visual capture.

Master’s thesis: Nonprobabilistic Topic Modeling of Short Texts with Distributed Representations Thesis research conducted in collaboration with UCLA, advised by Prof. Michael Möller and Prof. Andrea Bertozzi.

Bachelor’s thesis: Characterization of Pulse-Based Time-of-Flight Sensor Data Thesis grade: very good (1.0)