Delineating Dislocations
Event details
| Date | 22.05.2018 |
| Hour | 09:30 › 11:30 |
| Speaker | Okan Altingövde |
| Location | |
| Category | Conferences - Seminars |
EDIC candidacy exam
Exam president: Prof. Cécile Hebert
Thesis advisor: Prof. Pacal Fua
Co-examiner: Dr. Graham Knott
Abstract
Abstract—Dislocations in materials such as crystals carry
valuable information which may be incorporated to characterize
their topology and interaction mechanisms. Therefore, obtaining
accurate reconstruction of the 3D geometry of dislocations is
the key step for successfully analyzing material characteristics.
Delineation of dislocations is a challenging task due to ambiguities
in corresponding structures between images, variations in data
caused by acquisition processes and limited real-world data. In
this proposal, first we address three aspects, which are detection,
3D reconstruction and domain adaptation, of dislocation
delineation by discussing on related existing works: a deep
learning method for high accuracy edge detection, several deep
learning approaches for domain adaptations and a method to
reveal geometry between three images of same scene taken from
different views. Finally, we propose a research plan which stands
on previous works on aforementioned three core parts while
effectively combining them to achieve accurate delineation.
Background papers
Richer Convolutional Features for Edge Detection, by Liu Y, et al.
Deep-Learning Systems for Domain Adaptation in Computer Vision, by Venkateswara H., et al.
What Can Two Images Tell Us About a Third One?, by Faugeras O., Robert, L.
Exam president: Prof. Cécile Hebert
Thesis advisor: Prof. Pacal Fua
Co-examiner: Dr. Graham Knott
Abstract
Abstract—Dislocations in materials such as crystals carry
valuable information which may be incorporated to characterize
their topology and interaction mechanisms. Therefore, obtaining
accurate reconstruction of the 3D geometry of dislocations is
the key step for successfully analyzing material characteristics.
Delineation of dislocations is a challenging task due to ambiguities
in corresponding structures between images, variations in data
caused by acquisition processes and limited real-world data. In
this proposal, first we address three aspects, which are detection,
3D reconstruction and domain adaptation, of dislocation
delineation by discussing on related existing works: a deep
learning method for high accuracy edge detection, several deep
learning approaches for domain adaptations and a method to
reveal geometry between three images of same scene taken from
different views. Finally, we propose a research plan which stands
on previous works on aforementioned three core parts while
effectively combining them to achieve accurate delineation.
Background papers
Richer Convolutional Features for Edge Detection, by Liu Y, et al.
Deep-Learning Systems for Domain Adaptation in Computer Vision, by Venkateswara H., et al.
What Can Two Images Tell Us About a Third One?, by Faugeras O., Robert, L.
Practical information
- General public
- Free
Contact
- EDIC - [email protected]