Ukrainian Catholic University
Institutional Repository of the Ukrainian Catholic University ErUCUNot a member yet
3969 research outputs found
Sort by
Resilience of women whose partners are on the frontline in the context of attachment styles
No full textЗ початком повномасштабного вторгнення партнерки
військовослужбовців зустрілися з різноманітними стресорами. Вони не тільки
переживають тривалу розлуку з партнером, який знаходиться у постійної
небезпеці, але також залишаються сам на сам з побутовими, фінансовими,
соціальними забовʼязаннями, вихованням дітей. Окрім цього, вони також
переживають часті обстріли, відключення світла, адаптацію до нового місця
проживання будь то в Україні чи за кордоном.
Через високий рівень невизначеності, нестачу соціально-економічної
підтримки, нестабільну комунікацію з коханим, ця група населення має
підвищений ризик розвитку проблем з психічним здоровʼям, зокрема
депресії, тривожних розладів та зловживання ПАР (Donoho, C. J. et al., 2018;
Cole, R. F. et al., 2021). Підтримка резильєнтності жінок, які чекають своїх
партнерів з фронту, може бути протективним фактором проти труднощів з їх
ментальним здоровʼям (Erbes, C. R. et al., 2017). Тому важливо вивчати, що
впливає на їх резильєнтність, щоб мати можливість розробляти ефективні
програми допомоги.With the onset of the full-scale invasion, partners of military personnel have faced a variety of stressors. These women not only endure prolonged separation from their loved ones, who are constantly exposed to danger, but also bear the full burden of household, financial, and social responsibilities, including child-rearing. In addition, they experience frequent shelling, power outages, and the need to adapt to new living conditions—whether within Ukraine or abroad.
Due to high levels of uncertainty, limited socio-economic support, and unstable communication with their partners, this population is at increased risk of developing mental health issues, including depression, anxiety disorders, and substance abuse (Donoho, C. J. et al., 2018; Cole, R. F. et al., 2021). Supporting the resilience of women waiting for their partners to return from the frontline may serve as a protective factor against these mental health challenges (Erbes, C. R. et al., 2017). Therefore, it is crucial to study the factors that influence their resilience in order to develop effective support programs
The motif of suicide in the works of Ukrainian writers of the late 19th and early 20th centuries
No full textУ роботі досліджено мотив самогубства в українській літературі
кінця ХІХ – початку ХХ століття, де в центрі уваги – його художня
інтерпретація та аналіз зв’язку між творчістю та біографічних моментів авторів.
Праця включає дослідження історичного, соціокультурного та філософського
контекстів та впливу модерністських європейських течій. Розглядаються твори
та біографії М. Хвильового, В. Стефаника, О. Плюща та інших авторів, що
розкривають різні аспекти мотиву самогубства – від психологічних переживань
до ідейної інтерпретації. Дослідження може бути цікавим для філологів,
мистецтвознавців та тих, хто готовий зануритися в цю тему глибше.The paper explores the motive of suicide in Ukrainian literature of the late
19th and early 20th centuries, focusing on its artistic interpretation and analysis of the
connection between the work and the biographical moments of the authors. The work
includes a study of the historical, socio-cultural and philosophical contexts and the
influence of modernist European trends. The works and biographies of M.
Khvylovyy, V. Stefanyk, O. Plyushch and other authors are considered, revealing
various aspects of the motive of suicide – from psychological experiences to
ideological interpretation. The study may be interesting for philologists, art historians
and those who are ready to delve deeper into this topic
Interactive Object Segmentation in 3D Using 2D Foundation Models
No full textThis thesis explores capabilities of 2D foundational models for the task of interactive 3D object segmentation. Traditional 3D object segmentation often require extensive training dataset, and may struggle to generalize to other environments. Our work explores capabilities of usage of foundation 2D segmentation model to project segmentation output to 3D point cloud, enabling human-in-the-loop interaction for improved accuracy. We implement a SAM-based 3D segmentation pipeline and develop an interactive web tool for segmentation. Our experiments demonstrate that 2D foundation models can achieve competitive performance compared to other learning-based models, particularly with help of a human annotator. We discuss challenges such as evaluation prompt selection strategies and false positive segmentations and suggest future directions for that research. Our findings highlight the potential of 2D foundation models for flexible interactive 3D object segmentation without the need of additional training
Mitigating Dataset Bias in Human Pose Estimation via Novel View Synthesis
No full textHuman Pose Estimation (HPE) models often suffer from bias present in existing datasets, including limitations in camera viewpoints, controlled studio surroundings during the dataset’s creation, artifacts, and limited subjects, clothing, and backgrounds. All these problems affect models’ robustness and generalization abilities.
This thesis introduces the generation of a new synthetic dataset that aims to address these issues and remove the bias present in existing ones. Using a high-quality Diverse Neural Actor Repository for High-Fidelity Human-centric Rendering (DNA-Rendering) dataset, we process it with our pipeline, resulting in data from new realistic viewpoints taken from the distribution of cameras used in the Recovering Accurate 3D Human Pose in The Wild Using Inertial Measurement Unit and a Moving Camera (3DPW), captured in a realistic environment. We benchmark the performance of models trained on our dataset against the DNA-Rendering dataset with the Multi-Person Whole Body Human Mesh Recovery (Multi-HMR) framework, achieving better performance.
This dataset has great potential to improve the performance of models in the HPE field. The dataset generation pipeline and related code are made publicly available at https://github.com/YuriiKharabara/PoseBiasMitigation-NVS-Thesis.git
Comparison Analysis of Optimization Methods for the Location of Electric Vehicles Charging Stations in Urban Areas
No full textThe growing concern over environmental sustainability has accelerated the adoption of electric vehicles, driving the increasing demand for reliable and efficient EV charging infrastructure.
This study investigates and compares electric vehicle charging station placement optimization using three metaheuristic algorithms: genetic algorithm, particle swarm optimization, and simulated annealing. A demand simulation model based on a population distribution analysis is used to estimate the demand for charging stations. The optimization problem is formulated as a weighted sum of key objectives, including installation cost, spatial coverage, charging speed, and reducing negative impacts on the power grid.
Experiments were conducted using real road network data, and the algorithm’s performance was evaluated in terms of solution quality and convergence behavior. The results indicate that particle swarm optimization and genetic algorithm are highly effective in finding optimal solutions. In comparison, the simulated annealing algorithm proved to be less effective than the other mentioned methods for addressing this problem. The results highlight each method’s performance characteristics in the charging infrastructure optimization field. The implementation of the optimization methods and their testing can be found on the project’s GitHub repository
Розробка бізнес-моделі соціального підприємства «Центр взаємопідтримки ветеранів та ветеранок» ГО «Номад Лів»
No full textУ роботі аналізується соціальне підприємництво як інструмент
забезпечення фінансової стійкості неприбуткових організацій в контексті
підтримки ветеранів та ветеранок. Розроблено бізнес-модель соціального
підприємства на базі центру взаємопідтримки для ветеранів та ветеранок ГО
«Номад Лів». Представлено операційне та фінансове планування підприємства,
надано практичні рекомендації щодо його створення та розвитку. Результати
роботи можуть бути використані для впровадження нових соціальних ініціатив,
спрямованих на підтримку ветеранів, а також для розвитку соціального
підприємництва в Україні.
The paper analyses social entrepreneurship as a tool for ensuring the
financial sustainability of non-profit organisations in the context of supporting women
and men veterans. The business model of a social enterprise based on the mutual
support centre for women and men veterans of the NGO Nomad Live is developed.
Operational and financial planning of the enterprise is presented, and practical
recommendations for its creation and development are provided. The results of the
work can be used to introduce new social initiatives aimed at supporting veterans, as
well as to develop social entrepreneurship in Ukraine
“Erased”: the story of the disappearance of Ukrainian villages – a series of podcasts
No full textВсе більше людей в Україні стають свідомими щодо переходу на
українську мову, вивчення своєї історії та традицій. Повномасштабне
вторгнення в Україну 2022 року стало вирішальною точкою для багатьох, аби
споживати тільки україномовний контент. Усе це стимулювало ринок
україномовного ютубу, подкастів та іншого медіаконтенту. Зокрема, збільшився
попит на тему історії та культури. Сфера подкастингу відповіла на цей запит і
вже незабаром стали популярними такі подкасти як «Вулиця історії України»
від The Village, «Це було вже» від RADIO NV, «Без оголошення війни»,
«комік+історик» та інші.More and more people in Ukraine are becoming conscious of switching to the Ukrainian language, learning their history and traditions. The full-scale invasion of Ukraine in 2022 became a turning point for many, prompting them to consume only Ukrainian-language content. This shift has stimulated the growth of the Ukrainian-language YouTube space, podcasts, and other media content. In particular, there has been a growing demand for topics related to history and culture.
The podcasting sphere responded to this demand, and soon, podcasts such as “Ukraine’s History Street” by The Village, “It Has Already Happened” by RADIO NV, “Without a Declaration of War”, “Comic + Historian”, and others gained popularity
Development of a Swept-Volume LED Display for 3D Visualization
No full textThis thesis presents the development of a low-cost prototype for a swept-volume LED display, a form of volumetric display capable of rendering 3D imagery in real space without the requirement for special glasses or headsets. The system operates by rotating a 2D LED matrix to create a cylindrical display volume in which voxels are illuminated in synchronization with rotational angles, leveraging human visual persistence to produce a coherent 3D effect. The physiological, hardware, and software requirements for this display type are established to develop the working prototype.
The design integrates both hardware and software components, including a custom-built rotating platform, an infrared (IR) sensor for rotational synchronization, and a voxel-based rendering pipeline developed using Blender’s Python API. Experimental results confirm the system’s ability to render recognizable 3D shapes with minimal perceptual flicker. While limited by spatial resolution and static content, the prototype demonstrates the feasibility and potential of swept-volume display (SVD) technology for educational, artistic, and experimental applications. Future improvements are proposed in resolution, dynamic rendering capabilities, and power delivery systems
Visual Target Localization in Partially Known Environments
No full textFinding pre-defined targets in large outdoor environments from minimal prior data is an active problem in many modern applications like autonomous navigation, guidance systems, and medicine delivery. The most common solutions relying solely on spatial information face many issues: traditional GPS can be unreliable due to jamming or spoofing, and visual-inertial odometry accumulates error over time. Likewise, existing visual identification methods are based on either generic object detectors that require extensive training data and struggle generalizing to unseen targets, or template matchers that fail in textureless or repetitive outdoor environments. To tackle those issues, we propose a novel approach that fuses both spatial and visual information.
We focus on a specific scenario: real-time visual target localization in a large outdoor rural scene from bird’s-eye view (BEV) based on a small set of reference images that capture the target and its surroundings. To this end, we use only those methods that can work in partially known environments. For testing purposes, we develop a synthetic dataset generator that automatically crafts various scenes representing our use case.
After exhaustive testing under huge illumination and appearance changes, we discover that even in the most challenging scenarios, the proposed fusion approach does not degrade accuracy but only improves it. Furthermore, it outperforms a modern template matcher. The code, experiment results, and link to the generated dataset can be found on GitLab
Development of an Enhanced Ukrainian Text-to-Speech System
No full textText-to-speech (TTS) systems aim to convert written text into audible speech. The development of expressive and natural-sounding TTS systems for low-resource languages such as Ukrainian remains a significant challenge. This study explores the enhancement of Ukrainian TTS by evaluating and fine-tuning an existing TTS model, StyleTTS2, with a focus on the quality of adaptation to new voices while addressing the problem of limited data.
The study demonstrates that using precomputed speaker styles rather than sampling them with a diffusion model on each generation significantly improves the quality of generated utterances when adapting the model to new voices with as little as one hour of training data. In addition, the work introduces and evaluates a technique for embedding emotions into the synthesis process, allowing control over the emotional tones of the generated speech.
Finally, an evaluation is performed using the MOS (Mean Opinion Score) metric to determine the quality of the research findings regarding both naturalness and emotional expressiveness. We provide a repository with our code: https://github.com/nazar12314/ukrainian_tts.gi