About me

I'm Mira. I work in computer vision, used to spend most evenings in a floorball hall, and like being in the mountains whenever I can.

Research, via a sports hall

My first computer vision projects grew out of floorball. For my bachelor's and master's theses at CTU, I worked on tracking players across cameras. It was a way to connect something I was studying with a sport I already spent a lot of time thinking about.

At Porsche Engineering, I worked on Macan control-unit software and on the High Power Charging platform for Taycan and other electric vehicles. I've been a researcher in the Visual Recognition Group at CTU Prague since February 2019, and started my PhD in February 2023 with Jiří Matas. Since then I've visited Gerard Pons-Moll's group in Tübingen (December 2025–March 2026) and joined Qualcomm XR Labs in Amsterdam for a research internship (June–December 2026).

Human pose estimation was my starting point: understanding how people move and how to recover that reliably from images. My PhD grew from there into broader human body analysis, including detection, segmentation, pose and shape. I'm now extending that work into 3D reconstruction, spatial reasoning and how vision-language models understand people in their surroundings. At Qualcomm, I explore these questions in the context of personal AI assistants.

I'm drawn to the gap between a promising research result and something people can rely on. I think of it as the “last 10%”: the difficult cases and practical constraints that a strong benchmark score can hide. When I hear that a problem is solved, my first instinct is to stress-test that claim. What happens with an unfamiliar camera angle, a crowded scene or a body partly outside the frame?

For me, those failures are often where the next research question begins. I like developing new methods with a concrete use in mind, letting the needs of the people using a system shape what I investigate and how I evaluate it. The aim is to advance what works in practice as well as what is possible in research.

That interest has led to ProbPose, BBox-Mask-Pose and RePoGen, as well as datasets and evaluation metrics including CropCOCO, Extended OKS and OCHuman-Pose. I release code, models and demos, and build tools with the people using them: police analysts, infant-development researchers and floorball coaches. All projects.

Awards and service · Two-page CV · Earlier notes, 2022–2026

Teaching and supervision

I teach RPZ — Pattern Recognition and Machine Learning at CTU. The course covers statistical decision theory, perceptrons, AdaBoost, support vector machines and neural networks.

I also co-supervise research, including SAM-pose2seg, awarded Best Student Paper at CVWW 2026, and PC-CSE. Students I supervise have won prizes before even finishing their bachelor's degrees. Helping someone find their own research direction and write their first paper is a part of the work I particularly enjoy. I also coordinate our team's annotation work. I'm happy to discuss thesis ideas and research opportunities in my field.

  • For homework questions, please use the student forum.
  • I don't keep fixed consultation hours; email me to arrange a meeting.

A few years on the sidelines

I joined Sokol Královské Vinohrady in 2017 as an assistant coach. During Erasmus, I also coached FBK Olimpija in Ljubljana for a few months. Back in Prague, I took over as juniors' head coach, then became head coach of the men's Superliga team in 2024. We worked with a young team and a small budget, and reached the best results in the club's history through clear roles, shared effort and team spirit.

Those years taught me to enjoy both sides of a team: supporting someone else's direction as an assistant, and taking responsibility as head coach. I bring that same approach to research: contribute to the shared work, help others move forward, and lead when needed.

Some of the juniors I coached now play for the senior team. Watching that progression is what I remember most fondly. I stepped down in summer 2025 to focus on research, and stayed on as a consultant.

More about coaching, including the seasons, the people, and a podcast conversation in Czech.

Away from work

I enjoy mountains, travel, speciality coffee, psychology and epic fantasy. Steven Erikson’s Malazan Book of the Fallen is a particular favourite. The coffee hobby once became a small coffee-rating web app for our group at VRG.

Education

  • PhD (in progress) · Informatics

    2023 – present · Czech Technical University in Prague

    Robust Human Pose Estimation, supervised by prof. Jiří Matas. Expected completion 2026/27 (CV estimate). State doctoral exam passed with distinction, February 2025.

  • MSc · Computer Science — Artificial Intelligence, minors in Computer Vision and Cyber Security

    October 2020 – June 2022 · Czech Technical University in Prague

    Multi-object multi-view tracking; supervised by prof. Jiří Matas.

    Thesis

  • BSc · Cybernetics and Robotics

    2016–2020 · Czech Technical University in Prague

    Erasmus exchange at the University of Ljubljana, October 2018 – March 2019.

    Thesis

During my Erasmus exchange (October 2018–March 2019), I had my first computer vision research internship at the University of Ljubljana's Visual Cognitive Systems Laboratory, advised by prof. Matej Kristan.

Technical skills

These are my core skills — but really, with vibe coding I’m now picking up new languages and technologies much faster.

Python
Research implementations and deep learning, including training PMPose-h for more than ten days on eight GPUs using SLURM; daily automated testing at Porsche.
C / C++
Production automotive control-unit software, Plug & Charge for Porsche's charger, and an autonomous robot in a university course.
Frameworks & tools
PyTorch, OpenCV, SLURM, Git, SMPL, CUDA, Blender and MMCV.

Elsewhere

Interviews, features & coverage

For a compact overview, download my two-page CV.

Get in touch

I'm open to applied scientist and research engineer roles in computer vision, and selected consulting work. If you have a project or a question in mind, I'd be happy to hear from you.

mira.purkrabek@gmail.com