Master's thesis in Data & AI: Explainability for Deep Learning-Based Image Segmentation
Branche | Zie onder |
Dienstverband | Zie onder |
Uren | Zie onder |
Locatie | Veenendaal |
Salarisindicaties | 0-5.000 |
Opleidingsniveau | Zie onder |
Organisatie | Info Support |
Contactpersoon |
Info Support Nederland 0318552020 |
Informatie
- Computer Vision
- Explainable AI/XAI
- Object Detection
- Image Segmentation
- A challenging assignment within a practical environment
- € 1000 compensation, € 500 + lease car or € 600 + living space
- Professional guidance
- Courses aimed at your graduation period
- Support from our academic Research center at your disposal
- Two vacation days per month
- 65% Research
- 10% Analyze, design, realize
- 25% Documentation
Omschrijving
- Computer Vision
- Explainable AI/XAI
- Object Detection
- Image Segmentation
- A challenging assignment within a practical environment
- € 1000 compensation, € 500 + lease car or € 600 + living space
- Professional guidance
- Courses aimed at your graduation period
- Support from our academic Research center at your disposal
- Two vacation days per month
- 65% Research
- 10% Analyze, design, realize
- 25% Documentation
Functie eisen
Image segmentation refers to the detection and isolation of specific objects or regions of interest in an image. It can help to accurately extract the information that is relevant in an image, to use it for different purposes or even feed it to other models e.g. human action recognition, object tracking, background removal. When automated, image segmentation can be a powerful tool for several fields and applications, such as medical image processing, autonomous driving, agriculture, or video surveillance.
Image segmentation has typically been used in combination with (prior) object detection. Most state-of-the-art object detection and segmentation techniques are based on deep neural networks, which are black boxes by nature. While there has been relatively extensive research on explainable AI (XAI) techniques for object detection, there is little to no research available(1) for XAI applied to object segmentation.
Assignment
Having explainable state-of-the-art object segmentation would be extremely useful for many applications, especially those where high performance is critical but are subject to strict transparency regulations, such as those belonging to the medical field. Providing these sorts of explanations would be useful not only for regulation compliance and user confidence, but also for finding weak spots in segmentation models and eventually improving them.
The goal of this project would be to investigate the possibilities of XAI techniques applied to image segmentation, and come up with a novel XAI method for this purpose.
(1) An example of recent available research are these explainable medical image segmentation models: https://ieeexplore.ieee.org/abstract/document/9761664 & https://www.sciencedirect.com/science/article/abs/pii/S0010482522009398. However, while the explanations seem to be stable, the performance is not up to par with the state-of-the-art (plus the model is application-specific)