4D-YouTube: 4D Reconstruction from Casual Videos with Deep Space-Time Representations

• Applicant: Professor Dr. Daniel Cremers
• Subject Area: Image and Language Processing, Computer Graphics and Visualisation, Human Computer Interaction, Ubiquitous and Wearable Computing
• Term: since 2025
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 555335814

While camera-based reconstruction has evolved dramatically, the reconstruction of dynamic worlds from casual videos such as those on platforms like YouTube remains a significant open challenge. In this project, we will tackle this challenge in three stages: First, we will develop a framework for metric multi-view representation learning. Second, we will develop uncalibrarted visual SLAM techniques for static worlds that generalize existing SLAM methods so as to handle cameras with unknown calibration. And third, we will generalize these methods to the reconstruction of dynamic worlds where the 3D motion of objects in the world will be represented by means of ordinary differential equations.

• DFG Programme: Research Grants