Tomas Kerepecky

Versatile Academic and Leader: M.Sc. Graduate, PhD Student, M.A. Candidate, AI Freelancer, and Non-Profit & Church Leader

Hello! I am deeply engaged in academia and leadership, with a focus on advanced technology, education, and spirituality. Following my M.Sc. in Computational Physics, I am now pursuing a PhD in Image Processing and Artificial Intelligence at the Czech Academy of Sciences, linked to CTU in Prague. Additionally, I am a freelance AI specialist, especially in Large Language Models, working to demystify and master these complex AI systems. Alongside, I am studying for an M.A. in Practical Theology, concentrating on Leadership at TCM International Institute, Austria, merging my tech skills with a thorough understanding of "E10 Leadership" (see below).

In 2022, I achieved a milestone with the Fulbright-Masaryk award, enabling research at Washington University in St. Louis, US. Beyond academia, I have held educational and leadership roles, including as executive director of a non-profit Christian outdoor organization from 2015 to 2021 (resumed from 2024), missionary leader, church planter and a visiting teacher at New Hope Mission School in Bihar, India.

Download CV

Education and Research

  • 2021-2022

    Washington University in St. Louis, Missouri, USA

    Fulbright-Masaryk Award

  • 2019-now

    Department of Mathematics, Czech Technical University in Prague, Czechia

    PhD student in Mathematical Engineering. Focusing on Image processing.

  • 2018-now

    TCM International Institute, Vienna, Austria.

    MA student in Practical Theology concentrated on Leadership studies.

  • 2015-2019

    Department of Physical Electronics, Czech Technical University in Prague, Czechia

    M.Sc. degree in Computational physics. Master thesis: Inverse Compton scattering by laser-accelerated electrons.

    PhD student in Computational physics till 2019.

Professional and Research Interests

Image Processing & Computer Vision

My focus is on the intersection of technology and visual data, developing systems to process...

My focus is on the intersection of technology and visual data, developing systems to process, analyze, and interpret images and videos. This field has applications in medical imaging, autonomous vehicles, and facial recognition, emphasizing the enhancement of image analysis accuracy and adaptability in real-world scenarios.

Artificial Intelligence

I specialize in AI, aiming to replicate human intelligence in machines. My work covers...

I specialize in AI, aiming to replicate human intelligence in machines. My work covers deep learning in computer vision and natural language processing, focusing on algorithms that enable machines to learn, make decisions, and perform complex tasks. AI research drives innovation in healthcare, education, finance, and environmental management.

Prompt Engineering

As a prompt engineering specialist, I develop and optimize prompts for effective use of language models in...

As a prompt engineering specialist, I develop and optimize prompts for effective use of language models in various applications. This includes improving the capabilities of large language models in tasks like question answering and reasoning, ensuring their safety, and integrating them with external tools.

What is prompting?

E10 Leadership

My leadership philosophy is deeply rooted in the principles of servant leadership, a concept popularized by...�

My leadership philosophy is deeply rooted in the principles of servant leadership, a concept popularized by Robert Greenleaf but originally founded two millennia ago by Jesus of Nazareth, as described in the New Testament. This approach encourages us to Envision, Exemplify, Execute, Evaluate, and Enhance, as well as Engage, Encourage, Empower, Endorse, and Empathize with others. The E10 Leadership model embodies a commitment to serving first, from which leading naturally follows, aligning with Christian teachings and contemporary leadership practices.

TCM International Institute Non-profit Atleti v Akci, z.s.

Multidisciplinary leader at the intersection of technology, AI, education, and theology, researching to innovate and inspire.

Experiential Education

My interest in experiential education is grounded in the principles of Kolb's Experiential Learning...

My interest in experiential education is grounded in the principles of Kolb's Experiential Learning Theory, which posits that learning is most effective through a cycle of experiencing, reflecting, thinking, and acting. This approach resonates with my belief in the transformative power of hands-on experiences and reflective observation for personal growth and development. In this context, learning is seen as a continuous, life-long process, emphasizing the importance of actively engaging with and reflecting on experiences to gain deeper insights and practical knowledge.

Publications

Implicit Neural Representation for Image Demosaicking

2025 Digital Signal Processing: A Review Journal

We propose a novel approach to enhance image demosaicking algorithms using implicit neural representations (INR). Our method employs a multi-layer perceptron to encode RGB images, combining original Bayer measurements with an initial estimate from existing demosaicking methods to achieve superior reconstructions. A key innovation is the integration of two loss functions: a Bayer loss for fidelity to sensor data and a complementary loss that regularizes reconstruction using interpolated data from the initial estimate. This combination, along with INR's inherent ability to capture fine details, enables high-fidelity reconstructions that incorporate information from both sources. Furthermore, we demonstrate that INR can effectively correct artifacts in state-of-the-art demosaicking methods when input data diverge from the training distribution, such as in cases of noise or blur. This adaptability highlights the transformative potential of INR-based demosaicking, offering a robust solution to this challenging problem.

Read the article

Automated Actor Recognition in Video Content

2024 Chapter in Book: Data Science in Applications: Towards AI-driven Approaches.

This chapter presents an AI pipeline designed for automated recognition and analysis of actors in video content. The pipeline incorporates advanced methodologies in computer vision, allowing for a comprehensive analysis of actor presence and screen time across various video formats, such as movies, television shows, and surveillance footage.\newline\indent To evaluate the pipeline performance, we conducted extensive experiments using a carefully annotated test videos from a Czech TV show available for download. The evaluation criteria focus on precision, recall, mean absolute error metrics for actor recognition and screen time calculation under varying conditions. Additionally, we discuss challenges encountered during the pipeline development and consider its potential implications for the future of AI-driven content analysis and security surveillance.

Paper accepted, link soon

Inverse Problems in Image Restoration

2024 Chapter in Book: Extended Abstracts IPMS 2024 conference - "Inverse Problems: Modeling and Simulation"

This work addresses inverse problems in image restoration, focusing on recovering high-quality images from degraded observations, a critical task in fields like microscopy and digital photography. We examine both traditional variational methods and modern deep learning techniques, highlighting hybrid approaches that merge mathematical modeling with data-driven learning. Classical model-based methods use explicit regularization, like total variation, to incorporate prior knowledge and stabilize the inversion process. Meanwhile, deep learning approaches, both supervised and self-supervised, leverage implicit regularization, where network architectures capture and learn prior information from data. We present our recent advancements in this field and discuss the effectiveness of these complementary approaches in solving complex image restoration problems in theory and practice.

Paper accepted, link soon

STAR: Screen Time and Actor Recognition in Video Content

2024 The German Conference on Pattern Recognition (GCPR)

Accurately measuring the duration of actors' presence in videos is a challenging task that goes beyond actor recognition. We propose the STAR pipeline, the new model designed to analyze the time performers appear on screen across diverse video content, including movies and TV shows. The proposed model has been successfully deployed and tested by the Czech TV infrastructure provider. Our pipeline uses machine learning techniques for shot detection, face detection, tracking, recognition, and introduces a novel shot-based method for calculating screen time. We present extensive experiments proving the robustness and real-time performance of our approach. Alongside the pipeline, we introduce the STAR dataset to address the need for high-quality benchmarks in evaluating screen time models, now available for download.

Paper accepted, link soon

3D Non-separable Moment Invariants and Their Use in Neural Networks

2024 SN Computer Science Journal

Recognition of 3D objects is an important task in many bio-medical and industrial applications. The recognition algorithms should work regardless of a particular orientation of the object in the space. In this paper, we introduce new 3D rotation moment invariants, which are composed of non-separable Appell moments. We show that non-separable moments may outperform the separable ones in terms of recognition power and robustness thanks to a better distribution of their zero surfaces over the image space. We test the numerical properties and discrimination power of the proposed invariants on three real datasets-MRI images of human brain, 3D scans of statues, and confocal microscope images of worms. We show the robustness to resampling errors improved more than twice and the recognition rate increased by 2-10 % comparing to most common descriptors. In the last section, we show how these invariants can be used in state-of-the-art neural networks for image recognition. The proposed H-NeXtA architecture improved the recognition rate by 2-5 % over the current networks.

Read the article

NeRD: Neural field-based Demosaicking

2023 IEEE International Conference on Image Processing (ICIP)

We introduce NeRD, a new demosaicking method for generating full-color images from Bayer patterns. Our approach leverages advancements in neural fields to perform demosaicking by representing an image as a coordinate-based neural network with sine activation functions. The inputs to the network are spatial coordinates and a low-resolution Bayer pattern, while the outputs are the corresponding RGB values. An encoder network, which is a blend of ResNet and U-net, enhances the implicit neural representation of the image to improve its quality and ensure spatial consistency through prior learning. Our experimental results demonstrate that NeRD outperforms traditional and state-of-the-art CNN-based methods and significantly closes the gap to transformer-based methods.

Read the article

Real-Time Wheel Detection and Rim Classification in Automotive Production

2023 IEEE International Conference on Image Processing (ICIP)

This paper proposes a novel approach to real-time automatic rim detection, classification, and inspection by combining traditional computer vision and deep learning techniques. At the end of every automotive assembly line, a quality control process is carried out to identify any potential defects in the produced cars. Common yet hazardous defects are related, for example, to incorrectly mounted rims. Routine inspections are mostly conducted by human workers that are negatively affected by factors such as fatigue or distraction. We have designed a new prototype to validate whether all four wheels on a single car match in size and type. Additionally, we present three comprehensive open-source databases, CWD1500, WHEEL22, and RB600, for wheel, rim, and bolt detection, as well as rim classification, which are free-to-use for scientific purposes.

Read the article

Dual-Cycle: Self-Supervised Dual-View Fluorescence Microscopy Image Reconstruction using CycleGAN

2023 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)

Three-dimensional fluorescence microscopy often suffers from anisotropy, where the resolution along the axial direction is lower than that within the lateral imaging plane. We address this issue by presenting Dual-Cycle, a new framework for joint deconvolution and fusion of dual-view fluorescence images. Inspired by the recent Neuroclear method, Dual-Cycle is designed as a cycle-consistent generative network trained in a self-supervised fashion by combining a dual-view generator and prior-guided degradation model. We validate Dual-Cycle on both synthetic and real data showing its state-of-the-art performance without any external training data.

Read the article

D3Net: Joint Demosaicking, Deblurring and Deringing

2020 IEEE International Conference on Pattern Recognition (ICPR)

Images acquired with standard digital cameras have Bayer patterns and suffer from lens blur. A demosaicking step is implemented in every digital camera, yet blur often remains unattended due to computational cost and instability of deblurring algorithms. Linear methods, which are computationally less demanding, produce ringing artifacts in deblurred images. Complex non-linear deblurring methods avoid artifacts, however their complexity imply offline application after camera demosaicking, which leads to sub-optimal performance. In this work, we propose a joint demosaicking deblurring and deringing network with a light-weight architecture inspired by the alternating direction method of multipliers. The proposed network has a transparent and clear interpretation compared to other black-box data driven approaches. We experimentally validate its superiority over state-of-the-art demosaicking methods with offline deblurring.

Read the article

Iterative Wiener Filtering for Deconvolution with Ringing Artifact Suppression

2019 27th European Signal Processing Conference (EUSIPCO)

Sensor and lens blur degrade images acquired by digital cameras. Simple and fast removal of blur using linear filtering, such as Wiener filter, produces results that are not acceptable in most of the cases due to ringing artifacts close to image borders and around edges in the image. More elaborate deconvolution methods with non-smooth regularization, such as total variation, provide superior performance with less artifacts, however at a price of increased computational cost. We consider the alternating directions method of multipliers, which is a popular choice to solve such non-smooth convex problems, and show that individual steps of the method can be decomposed to simple filtering and element-wise operations. Filtering is performed with two sets of filters, called restoration and update filters, which are learned for the given type of blur and noise level with two different learning methods. The proposed deconvolution algorithm is implemented in the spatial domain and can be easily extended to include other restoration tasks such as demosaicing and super-resolution. Experiments demonstrate performance of the algorithm with respect to the size of learned filters, number of iterations, noise level and type of blur.

Read the article

Inverse Compton scattering by laser-accelerated electrons

Czech Technical University in Prague. Computing and Information Centre, 2017

This thesis deals with the study of X- and -radiation during the interaction of relativistic electrons with the intense electromagnetic field. This mechanism is called inverse Compton scattering. For the purpose of examining the properties of radiation from inverse Compton scattering, a new COCO code has been implemented. Radiation spectrum is computed through the use of fast Fourier transform of the radiation field of electrons.

Read the thesis

Teaching and AI

Contact.

kerepecky@utia.cas.cz +420 266 052 864
  • Filip Sroubek's Lab,
  • Department of Image Processing,
  • Institute of Information Theory and Automation,
  • Czech Academy of Sciences.