Simona Kocour

I am an ELLIS PhD student at CTU Prague, supervised by Torsten Sattler and Siyu Tang at ETH Zurich. My research focuses on spatial intelligence and representations in vision foundation models. I am particularly interested in how models represent objects, geometry, identity, motion, and interactions, and how these representations can support world models and embodied agents.

I received my Bachelors and Masters degree in Mathematics from Masaryk University. After the Masters, I worked as a Data Scientist, and then as a Machine Learning Researcher for 3 years.

Email  |  CV  |  LinkedIn

headshot
Current Research

Object Identity in Vision Foundation Models
I study how physical object identity is represented across layers and training objectives, including the trade-off between viewpoint invariance and distinguishing look-alike objects.

Dynamic Spatial Intelligence
I’m extending object-centric 3D representations toward persistent scene understanding, with a focus on maintaining object identity, spatial relations, motion, and interactions over time.

Publications
Qualitative example of structured 3D scene representation: labeled 3D bounding boxes for car, truck, post, and sign over a point cloud reconstruction Structured 3D Scene Representations Improve Spatial Reasoning Across Benchmarks
Simona Kocour, Yutong Chen, Torsten Sattler, Sergey Prokudin, Siyu Tang
Under review
DescOver: Descriptor Overlap Leaks the Image Behind Features
Simona Kocour, Torsten Sattler, Zuzana Kukelova, Marc Pollefeys, Pablo Speciale
Under review
Remove360 teaser: image renders, SAM segments, semantic segmentation, and depth, before and after table removal Remove360: Benchmarking Residuals After Object Removal in 3D Gaussian Splatting
Simona Kocour, Assia Benbihi, Torsten Sattler
ECCV Workshop on Structure-from-Motion in the Age of Deep Learning (SFM-DL), 2026
Paper  /  Code  /  Dataset
News
09/2026Our paper "Remove360: Benchmarking Residuals After Object Removal in 3D Gaussian Splatting" was accepted at the ECCV 2026 Workshop on Structure-from-Motion in the Age of Deep Learning (SFM-DL).
08/2025 – 08/2026Research visit at ETH Zurich. I was part of the Visual Learning Group, working with Prof. Dr. Siyu Tang and Dr. Sergey Prokudin on Spatial Intelligence in 3D.
10/2024Volunteering at ECCV 2024 in Milan, Italy.
09/2024I got accepted to the ELLIS PhD Program. My PhD is part of the pan-European AI network of excellence.
06/2024Attended Symposium on Geometry Processing at MIT, Cambridge, MA, USA.
04/2024Career Update: Started PhD journey at CTU.
07/2023Attended AI4Health Summer School in Paris, France.
11/2020 – 03/2024I worked as an NLP Machine Learning Researcher at Emplifi in Prague, Czech Republic.
06/2020Graduated Master in Mathematics from Masaryk University in Brno, Czech Republic.

template adapted from this awesome website