Register now After registration you will be able to apply for this opportunity online.
This opportunity is not published. No applications will be accepted.
Master Thesis Project on Learning based Map Representations for Visual 3D Scene Reconstruction
Many common map representations for 3D reconstruction discretize the represented volume into a grid of voxels or the surface into a large set of points. In this Master thesis project, a deep learning based 3D map representation shall be developed that overcomes shortcomings of such discretizations.
Keywords: 3D scene reconstruction, 3D scanning, 3D map representation, deep learning
3D scene reconstruction is an important prerequisite for many applications in augmented reality and robotics. Visual sensors such as monocular or RGB-D cameras are specifically interesting for this purpose, since they are small, light-weight, low-power consuming and also typically less costly than other sensors such as LiDAR.
Common map representations for 3D scene reconstruction usually ultimately either discretize the represented volume into a grid of voxels, or the scanned surface into a large set of points. In this project, a deep learning based 3D map representation for scene reconstruction shall be developed that overcomes the shortcomings of such discretizations by learning encodings of the 3D structure. Details about possible approaches to reach this goal are subject to discussion with prospective candidates.
We are looking for an outstanding Master of Science student in computer science, physics, mathematics, or electrical engineering from any university who is eager to do a Master thesis on this project. The thesis project will be conducted at the Max Planck Institute for Intelligent Systems (MPI-IS) located in Tübingen. Note that depending on their study programme regulations, students interested in writing a Master thesis at the MPI might have to find a supervisor/examiner at their home institution who is willing to support the Master thesis.
**Prerequisites**
High motivation, excellent skills in computer science, a solid background in mathematics and hands-on experience with deep learning are prerequisites. Furthermore, good software engineering skills in C/C++ and Python are required. A background in computer vision, specifically in structure-from-motion and SLAM, and prior programming experience in TensorFlow is a plus.
**Embodied Vision Group**
The project will be carried out at the Max Planck Institute for Intelligent Systems (MPI-IS) located in Tübingen within the Embodied Vision Group headed by Dr. Joerg Stueckler. The group investigates fundamentals of embodied intelligent agents such as robots that learn to perceive and act within their environment (https://ev.is.mpg.de).
3D scene reconstruction is an important prerequisite for many applications in augmented reality and robotics. Visual sensors such as monocular or RGB-D cameras are specifically interesting for this purpose, since they are small, light-weight, low-power consuming and also typically less costly than other sensors such as LiDAR.
Common map representations for 3D scene reconstruction usually ultimately either discretize the represented volume into a grid of voxels, or the scanned surface into a large set of points. In this project, a deep learning based 3D map representation for scene reconstruction shall be developed that overcomes the shortcomings of such discretizations by learning encodings of the 3D structure. Details about possible approaches to reach this goal are subject to discussion with prospective candidates.
We are looking for an outstanding Master of Science student in computer science, physics, mathematics, or electrical engineering from any university who is eager to do a Master thesis on this project. The thesis project will be conducted at the Max Planck Institute for Intelligent Systems (MPI-IS) located in Tübingen. Note that depending on their study programme regulations, students interested in writing a Master thesis at the MPI might have to find a supervisor/examiner at their home institution who is willing to support the Master thesis.
**Prerequisites** High motivation, excellent skills in computer science, a solid background in mathematics and hands-on experience with deep learning are prerequisites. Furthermore, good software engineering skills in C/C++ and Python are required. A background in computer vision, specifically in structure-from-motion and SLAM, and prior programming experience in TensorFlow is a plus.
**Embodied Vision Group** The project will be carried out at the Max Planck Institute for Intelligent Systems (MPI-IS) located in Tübingen within the Embodied Vision Group headed by Dr. Joerg Stueckler. The group investigates fundamentals of embodied intelligent agents such as robots that learn to perceive and act within their environment (https://ev.is.mpg.de).
Do not hesitate to contact us (see contact details below) if you are interested in doing your Master thesis on this project. Applications should be sent in a single pdf (max. 10MB) per email and include a CV, a short motivation letter (why are you interested in this project?), current transcripts of BSc/MSc studies, and optionally other documentation helpful to evaluate your background.
Do not hesitate to contact us (see contact details below) if you are interested in doing your Master thesis on this project. Applications should be sent in a single pdf (max. 10MB) per email and include a CV, a short motivation letter (why are you interested in this project?), current transcripts of BSc/MSc studies, and optionally other documentation helpful to evaluate your background.
Dr. Joerg Stueckler | joerg.stueckler@tuebingen.mpg.de
MPI for Intelligent Systems, Max-Planck-Ring 4, 72076 Tübingen, Germany.
+49 (0) 7071-601 385 | http://is.mpg.de/person/jstueckler
Dr. Joerg Stueckler | joerg.stueckler@tuebingen.mpg.de MPI for Intelligent Systems, Max-Planck-Ring 4, 72076 Tübingen, Germany. +49 (0) 7071-601 385 | http://is.mpg.de/person/jstueckler