The Machine Learning Lab

Our R&D projects and publications

The Machine Learning Lab was founded in 2018 at the Faculty of Applied Sciences, Ukrainian Catholic University (UCU) in Lviv. Current research covers machine learning, computer vision, reinforcement learning. We are actively working on new publications, and our center has also well-established connections with industry.


Weakly supervised semantic segmentation

To create models which can effectively segment images, the immense datasets of labeled data are required. It is often very time consuming. When labeling medical data an expert has to spend a lot of time and use his expertise which is very expensive.

In order to decrease the resources spent on labeling while preserving the quality of the models, we develop a weakly-supervised approach for segmentation of different types of images. We test our method and show results on several distant datasets, including regular real-world objects (such as cars, planes, trees etc.), and on medical data (chest X-rays). We develop a solution which is capable to segment objects with a limited supervision.

ModelicaGym Toolbox for Applying Reinforcement Learning to Modelica Models

ModelicaGym toolbox was developed to employ Reinforcement Learning (RL) for solving optimization and control tasks in Modelica models. The developed tool allows connecting models using Functional Mock-up Interface (FMI) to OpenAI Gym toolkit in order to exploit Modelica equation-based modeling and co-simulation together with RL algorithms as a functionality of the tools correspondingly. Thus, ModelicaGym facilitates fast and convenient development of RL algorithms and their comparison when solving optimal control problem for Modelica dynamic models. The toolbox functionality validation is performed on Cart-Pole balancing problem.

Authors: Oleh Lukianykhin, Tetiana Bogodorova

Minimal Solvers for Rectifying from Radially-Distorted Scales and Change of Scales

The project introduces the first minimal solvers that jointly estimate lens distortion and affine rectification from the image of rigidly-transformed coplanar features. The solvers work on scenes without straight lines and, in general, relax strong assumptions about scene content made by the state of the art. The proposed solvers use the affine invariant that coplanar repeats have the same scale in rectified space. Synthetic and real-image experiments confirm that the proposed solvers demonstrate superior robustness to noise compared to the state of the art. Accurate rectifications on imagery taken with narrow to fisheye field-of-view lenses demonstrate the wide applicability of the proposed method. The method is fully automatic.

Authors: James Pritts, Zuzana Kukelova, Viktor Larsson, Yaroslava Lochman, Ondřej Chum

Face Recognition system

We developed a loyalty system for retail based on the face recognition. A user has to register via mobile phone and take one photo. After that based on the angularly discriminative features we build engine for face recognition. Now when the registered client comes to the shop they will be recognised and the shop assistant will immediately get all the information about the client's tastes so they can recommend better items.

Authors: Oles Dobosevych, Anton Tarasov, Matt Kovtun, Oleh Smolkin, Mykhailo Ivankiv

Ukrainian style MNIST

Modified National Institute of Standards and Technology dataset (MNIST dataset) of handwritten digits is the most known dataset that is widely used as a benchmark for validating various ideas in Machine Learning. We present a newly created dataset of 32 handwritten Ukrainian letters, which is divided into 72 different style subclasses, with 2000 examples in each class. We also suggest a recognition model for these symbols and explain why approaches working well for MNIST dataset do not succeed in our case. Finally, we discuss several real-world applications of our model that can help to save paper, time and money.

Authors: Oles Dobosevych, Matvii Kovtun

Menu generator

We developed a system that allows restaurant owners to customise menus for their restaurants in a few clicks and FFS owners to collect and analyse data about dishes that restaurants prefer to choose.

Authors: Kostiantyn Liepieshov, Vladyslav Ursul

Low cost system to record punch press parameters

We developed an IoT solution to automatically count the number of overall press strikes as well as granulate this data to individual molds with minimum operator intervention.

Authors: Dzvenymyra Yarish, Yurii Laba, Yurii Stasinchuk, Oleksandr Pryhoda, Bogdan Petryshak


J. Pritts, Z. Kukelova, V. Larsson, Y. Lochman, O. Chum. Minimal Solvers for Rectifying from Conjugate Translations. arXiv preprint, 2019


O. Kupyn, D. Pranchuk. Fast and Efficient Model for Real-Time Tiger Detection In The Wild. arXiv preprint, 2019


O. Kupyn, Tetiana Martyniuk, Junru Wu, Zhangyang Wang. DeblurGAN-v2: Deblurring (Orders-of-Magnitude) Faster and Better. arXiv preprint, accepted for ICCV 2019


Irynei Baran, O. Kupyn, A. Kravchenko. Safe Augmentation: Learning Task-Specific Transformations from Data. arXiv preprint, accepted for WACV 2020


J. Pritts, Z. Kukelova, V. Larsson, Y. Lochman, O. Chum. Minimal Solvers for Rectifying from Radially-Distorted Scales and Change of Scales. arXiv preprint, accepted for publication in IJCV, 2019


O. Lukianykhin, T. Bogodorova. ModelicaGym: Applying Reinforcement Learning to Modelica Models. arXiv preprint, accepted for publication in EOOLT 2019


O. Kupyn, V. Budzan, M. Mykhailych, D. Mishkin, J. Matas. DeblurGAN: Blind Motion Deblurring Using Conditional Adversarial Networks, IEEE CVPR, Salt Lake City, USA, June 2018
Publication, Poster


M. Mykhailych. Application of Generative Neural Models for Style Transfer Learning in Fashion, Master Thesis, 2018


A. Stehnii. Generation of code from text description with syntactic parsing and Tree2Tree model, Master Thesis, 2018