Projects / Programmes source: ARIS

MV4.0 - Data-driven framework for development of machine-vision solutions

Research activity

Code Science Field Subfield
2.07.00  Engineering sciences and technologies  Computer science and informatics   

Code Science Field
1.02  Natural Sciences  Computer and information sciences 
Deep learning, machine vision, Industry 4.0, computer vision, supervised learning, unsupervised learning, visual learning, 6DOF object pose detection, visual inspection, surface-defect detection, automation, digitalisation
Evaluation (rules)
source: COBISS
Data for the last 5 years (citations for the last 10 years) on April 13, 2024; A3 for period 2018-2022
Data for ARIS tenders ( 04.04.2019 – Programme tender, archive )
Database Linked records Citations Pure citations Average pure citations
WoS  166  7,972  7,623  45.92 
Scopus  246  12,082  11,505  46.77 
Researchers (8)
no. Code Name and surname Research area Role Period No. of publicationsNo. of publications
1.  29381  PhD Luka Čehovin Zajc  Computer science and informatics  Researcher  2021 - 2024  124 
2.  55044  Matej Dobrevski  Computer science and informatics  Researcher  2021 - 2024  13 
3.  30155  PhD Matej Kristan  Computer science and informatics  Researcher  2021 - 2024  323 
4.  05896  PhD Aleš Leonardis  Computer science and informatics  Researcher  2021 - 2024  455 
5.  50843  Jon Natanael Muhovič  Computer science and informatics  Researcher  2021 - 2024  23 
6.  57319  Marko Rus  Computer science and informatics  Researcher  2022 - 2024 
7.  18198  PhD Danijel Skočaj  Computer science and informatics  Head  2021 - 2024  309 
8.  34398  PhD Domen Tabernik  Computer science and informatics  Researcher  2021 - 2024  50 
Organisations (1)
no. Code Research organisation City Registration number No. of publicationsNo. of publications
1.  1539  University of Ljubljana, Faculty of Computer and Information Science  Ljubljana  1627023  16,226 
The Industry 4.0 paradigm has made machine vision systems a necessity in modern industrial facilities. They enable digitalisation of perception and provide valuable visual information about the production processes that facilitate intelligent information extraction and decision making. The main functional objective of the proposed project is to change the development and deployment pipeline of machine vision systems. Our goal is to shift the predominant paradigm of developing hand-engineered specific solutions into the direction of data-driven learning-based design and development that would enable more general, efficient, flexible and economical development, deployment and maintenance of machine vision systems. To this end, the main applied objective of the project is to develop a software framework that would enable such kind of development with as little and undemanding involvement of a human operator as possible, minimising the requirement for manual data acquisition and annotation by highly automating the entire process of data preparation and model training. This goal requires solving a number of scientific problems, such as developing core deep-learning methods and procedures for iterative, active, robust, few-shot, weakly-, self-, and un-supervised learning of visual models that will reach the required performance with minimal (manually annotated) training images. The challenges will be tackled from three directions. (1) We will develop a procedure for automated image synthesis augmented with synthetic-to-real domain adaptation for generating training images to facilitate efficient supervised learning with unlimited data without requiring significant human labour. (2) We will also develop methods for efficient learning from a limited amount of annotated data by exploring general visual information utilised for few-shot learning. Training data is often noisy in practice; we will thus develop robust and active learning techniques for filtering erroneous training annotations and for actively selecting the training data requiring annotation to further reduce the human work. (3) For the problem domains, in which unannotated data is abundantly available, we will develop methods for unsupervised and self-supervised learning without data annotation. Although the principles that we will be developing are quite general, we will focus our work on a specific problem domain of industrial machine vision. The developed principles will be applied on two very common problems in industrial settings: (i) 6DOF object pose detection, which is a very important perceptual task in robotic grasping applications, and (ii) surface-defect detection, frequently encountered in visual inspection systems. We will demonstrate the technological advancements of the developed approaches in a final demonstrator at TRL level 4. We expect to go beyond the state-of-the-art on these research topics. Beside significant scientific contributions, the developed solutions will also have a substantial application value. Given that the entire computer vision field has been conquered by more effective deep-learning methods, the data-driven deep-learning-based solutions will inevitably prevail in the rather conservative machine vision sector as well. It is essential that the solution providers adapt and extend their activities with such state-of-the-art approaches. These approaches, however, require and allow a different software development process. In this project, we will develop a prototype of a general framework for rapid development and deployment of machine vision applications that will follow this paradigm. By introducing this paradigm in their solutions, the providers are expected to gain the upper hand in the highly competitive market. This project will help them to achieve this goal.
Views history