Spring Meeting 2024: Recent Advances of Deep Learning in Computer Vision: Innovations & Applications
Deep Learning in Computer Vision
The NVPHBV Spring Meeting 2024 will be about Recent advances of Deep Learning in Computer Vision: Innovations and applications.
In an inspiring day, we are going to explore resent advances with invited keynote speakers and we will have a series of short abstract presentations. The winner of the best-abstract award will be announced at the end of the meeting. As always there will be ample time to connect with all attendees from industry and academia.
Keynotes:
- 10:45 – 11:30 Prof. dr. Jiri Kosinka (University of Groningen): A (not so) deep look into image colourisation and shape segmentation
- 14:00 – 14:45 Laurens Hogeweg (Intel): Why you should work in AI for biodiversity?
- 14:45 – 15:15 Dr. Matias Valdenegro Toro (University of Groningen): Bayesian Deep Learning for Safe Computer Vision
Abstract presentations:
- 12:00 – 12:15 Melissa Tijink et al: Automatic Wagon Code Identification using Computer Vision
- 12:15 – 12:30 Dovile Juodelyte et al: Source Matters: Source Dataset Impact on Model Robustness in Medical Imaging
- 12:30 – 12:45 Yeşim Hekim Tanç et al: Classification of in situ EV Battery cells with high-resolution CT using deep learning
- 12:45 – 13:00 Florens de Wit: Limitations and practical considerations for personal data usage for AI development and testing
Hourly Schedule
Agenda
- 10:00 - 10:25
- Welcome - Coffee & Tea
- 10:25 - 10:30
- Welcome from NVPHBV president
-
Speakers:
Prof. Dr. Clarissa Sánchez
- 10:30 - 10:45
- Introduction by hosts
-
Speakers:
Prof. dr. ir. Peter van Ooijen, Prof.dr. George Azzopardi
- 10:45 - 11:30
- Keynote - A (not so) deep look into image colourisation and shape segmentation
- From its humble origins approximately 60 years ago, computer vision has developed into a mature interdisciplinary field that tackles challenging problems revolving around obtaining high-level understanding from digital images, videos, shapes, etc. Recent deep learning advances have transformed the field, allowing for solving problems that were previously beyond reach. We will focus on two problems where deep learning comes to the rescue: segmenting (families of) 3D shapes and colouring greyscale images, with image vectorisation linking these two challenges.
-
Speakers:
Prof. dr. Jiri Kosinka
- 11:30 - 12:00
- Break - Coffee/Tea & Poster Session
- 12:00 - 13:00
- Abstract Presentations
-
Speakers:
Prof. dr. Marcel Breeuwer
- 13:00 - 14:00
- Lunch & Poster Session
- 13:45 - 14:00
- Member Meeting (NVPHBV members only)
-
Speakers:
Prof. Dr. Clarissa Sánchez
- 14:00 - 14:45
- Keynote - Why you should work in AI for biodiversity?
- The natural world, including biodiversity, provides a challenging playing ground for developing machine learning techniques that scale and apply to the real world. This talk discusses machine learning research applied to biodiversity performed at Naturalis Biodiversity Center and Intel. It discusses how this research is applied in practice to support large scale automatic monitoring of biodiversity and help citizen scientists collect biodiversity data and become engaged with protecting nature. The talk ends with a discussion of future developments including the rapid development of large scale models and edge AI.
-
Speakers:
Laurens Hogeweg
- 14:45 - 15:15
- Keynote - Bayesian Deep Learning for Safe Computer Vision
- What if we train a model to classify dogs and cats, but it is later tested with an image of a human? Generally the model will output either dog or cat, and has no ability to signal that the image has no class that it can recognize. This is because classical neural networks do not contain ways to estimate their own uncertainty (so called epistemic uncertainty), and this has practical consequences for the use of these models, like safety when cooperating with humans, autonomous systems like robots, and computer vision systems. A possible solution is the bayesian neural network. In this talk I will cover the basic concepts of bayesian neural networks, and how they can help us to produce safer models, including explainable AI and computer vision.
-
Speakers:
Dr. Matias Valdenegro Toro
- 15:15 - 15:30
- Abstract Award Ceremony
-
Speakers:
Prof. Dr. Clarissa Sánchez
- 15:30 - 16:30
- Drinks & Networking
Speakers
-
Prof. dr. Jiri KosinkaAssociate Professor at University of Groningen
Jiří Kosinka is an Associate Professor at the Bernoulli Institute of the Faculty of Science and Engineering, University of Groningen, where he leads the Scientific Visualization and Computer Graphics research group. His interests include topics in the area of visual computing, with particular emphasis on geometric modelling, computer graphics, and image vectorization. He currently serves as Associate Editor for two Elsevier journals, Computer Aided Design and Graphical Models. He has been involved in organising several conferences in his field, such as the SIAM Conference on Computational Geometric Design as programme co-chair in 2021 and conference co-chair in 2023. He has co-authored over 100 scientific publications and served on more than 40 international programme committees.
-
Laurens HogewegMachine Learning Researcher at Intel Corporation
Laurens Hogeweg is a highly accomplished professional with a diverse background in machine learning and deep learning. He is currently associated with Intel Benelux BV and Naturalis Biodiversity Center.
Laurens Hogeweg’s work at Intel Benelux BV is primarily focused on the development of out-of-distribution (OOD) detection methods for large-scale hierarchical classification. His work is particularly relevant to the species recognition task in images, which involves dealing with large databases, a large number of fine-grained hierarchical classes, severe class imbalance, and varying image quality.
In addition to his role at Intel, Laurens Hogeweg also contributes to the Naturalis Biodiversity Center. His work here involves the application of deep learning techniques for species recognition from images. He has been involved in research that investigates how the amount of training data influences the performance of species recognition models for various taxa.
Prior to his current roles, Laurens Hogeweg obtained a PhD in machine learning of medical images at Radboud University Nijmegen. After obtaining his PhD, he broadened his focus to include other areas of machine learning and deep learning.
Laurens Hogeweg’s work is characterized by his ability to apply advanced machine learning techniques to practical problems. His contributions to the field of machine learning, particularly in the context of species recognition and hierarchical classification, are noteworthy.
-
Dr. Matias Valdenegro ToroAssistant Professor for Machine Learning at University of Groningen
Dr. Matias Valdenegro is Assistant Professor for Machine Learning at the Department of Artificial Intelligence, Bernoulli Institute, University of Groningen since 2022. He studied Computer Science, Autonomous Systems, and Electrical Engineering in Chile, Germany, and Scotland. As a Researcher at the German Research Center for Artificial Intelligence in Bremen he conducted research in Computer Vision and Uncertainty Quantification from 2018 to 2022.
He also has 5 years of professional experience as a Software Engineer in Latinoamerica.
His research interests are Uncertainty in Machine Learning, Explainable AI, Robot Perception, and their applications to Robotics and AI Safety. Matias is part of the LatinX in AI community, organizing affinity workshops and networking events, and has received many best reviewer awards at top machine learning conferences (including NeurIPS, ICML, ICLR), and has authored over 50 peer-reviewed publications.