Register now After registration you will be able to apply for this opportunity online.
This opportunity is not published. No applications will be accepted.
Leveraging Neural Scene Representations for Large-Scale Localization
Many real-world applications for Augmented Reality (AR) heavily rely on a stable and accurate localization algorithm.
AR content pinned to a physical environment can only be seen at the right location by the users once they are correctly localized into that environment, and for multiple users to share an AR experience, all those users must be co-localized w.r.t. each other into the same digital space.
While current large-scale localization approaches, particularly those targeting real-world applications, still heavily rely on explicit, geometric scene representations (e.g. graph-based maps), recent advances in the field of computer vision have shown impressive results in using machine learning techniques for implicitly representing scenes.
Keywords: AR
In this project, we want to explore the applicability of current SOTA methods in neural scene representation to large-scale localization, and investigate their potential to improve such approaches. We will investigate how well different SOTA techniques can represent the environment in large and complex AR scenarios. Based on those findings, we will identify and explore directions on how those techniques can be used to improve parts of Magic Leap's algorithmic pipeline for large-scale localization.
You are:
A motivated Master's student with an excellent background in Computer Vision and Machine Learning, as well as a solid knowledge of 3D geometry. Knowledge of C++, Python and Pytorch are mandatory. Additional background in localization, mapping and SLAM are a plus.
In this project, we want to explore the applicability of current SOTA methods in neural scene representation to large-scale localization, and investigate their potential to improve such approaches. We will investigate how well different SOTA techniques can represent the environment in large and complex AR scenarios. Based on those findings, we will identify and explore directions on how those techniques can be used to improve parts of Magic Leap's algorithmic pipeline for large-scale localization.
You are:
A motivated Master's student with an excellent background in Computer Vision and Machine Learning, as well as a solid knowledge of 3D geometry. Knowledge of C++, Python and Pytorch are mandatory. Additional background in localization, mapping and SLAM are a plus.
Master Thesis Neural Scene Representations.docx.pdf
