Public defence in Computer Science, M.Sc.(Tech) Shuzhe Wang

Title of the thesis: Deep Learning Methods for Point Matching, Visual Localization and 3D Reconstruction

Doctoral Student: Shuzhe Wang
Opponent: Assistant Professor Viktor Larsson, Lund University, Sweden
Custos: Associate Professor Juho Kannala, Aalto University School of Science, Department of Computer Science

This doctoral thesis explores end-to-end, data-driven approaches for 3D computer vision tasks, such as image matching, visual localization, and 3D reconstruction. The research addresses the limitations of traditional 3D vision pipelines, which are often complex and require substantial engineering, by developing fully data-driven methods using modern deep learning techniques.

The main result of this doctoral study is the development of novel, end-to-end deep learning pipelines that achieve or exceed the performance of traditional methods in multiple 3D tasks in terms of accuracy and scalability. This thesis demonstrates the potential for deep learning to generalize better across diverse 3D scenarios, offering more dynamic and robust solutions compared to conventional approaches.

The research is highly relevant to the broader field of computer vision, contributing to advancements in areas like augmented reality, robotics, and autonomous driving, where accurate 3D scene understanding is crucial. The proposed approaches reduce the need for extensive data storage space and reduce computational costs, making them suitable for real-time applications. More importantly, the thesis proposes an idea of a 3D foundation model, which opens new possibilities for efficiently deploying 3D vision solutions in applications that require high-quality spatial perception and interaction.

Key words: 3D computer vision, Machine Learning, Data-driven

Thesis available for public display 10 days prior to the defence at: https://aaltodoc.aalto.fi/doc_public/eonly/riiputus/ 

Contact information:

Doctoral theses of the School of Science: https://aaltodoc.aalto.fi/handle/123456789/52