
Master's Thesis — Monocular Dynamic Motion Capture
A Regression-optimization hybrid approach for monocular dynamic motion capture.

I’m a research engineer at Hochschule Luzern working on human-centric computer vision, including dynamic 3D representations, metric deep learning, and motion recovery, under the advise of Prof. Aljosa Smolic. I like turning research ideas into fast, practical tools that people can use.
Before this, I completed a MSc in Machine Learning at KTH Royal Institute of Technology, where I worked with Prof. Hedvig Kjellström. Earlier, I earned a Diploma in Electrical & Computer Engineering from Aristotle University of Thessaloniki, where I worked with Prof. Pericles Mitkas. I’m based in Rotkreuz, Switzerland, while my hometown is Thessaloniki, Greece.
Outside of work, I used to make electronic music and do graphic design. These days I relax with freelance web development, my go-to antistressant. When I’m away from the keyboard, I’m usually on a bike in the mountains or forests.

![[Re] Masked Autoencoders Are Small Scale Vision Learners: A Reproduction Under Resource Constraints](/media/publications/charisoudis2023smae/960.jpg)

A Regression-optimization hybrid approach for monocular dynamic motion capture.

Developed GAN pipelines for realistic clothing transformations; compared pix2pix, CycleGAN, StyleGAN, MUNIT.

Capturing and rendering of humans in motion for immersive reality.

GANs and flow-based generative models for vision tasks.

Neural networks for vision and computational acoustics; deep RL.

AI techniques for robot systems with strong perception.

High-quality, stable code guided by extensive unit tests.

Fan of Next.js stacks. Experience with PHP/Laravel.