4D Reconstruction and Editing
Project Description
The project aims to explore methods for the reconstruction and editing of dynamic 3D scenes (4D: spatial + temporal dimensions). Recent advances in neural rendering and video generation have opened new possibilities for creating editable 4D content from 2D video inputs. However, challenges remain in achieving temporal consistency, handling occlusions, and enabling user-friendly editing.
Supervisor
XU, Dan
Quota
1
Course type
UROP1100
UROP3200
Applicant's Roles
• Reconstruct 4D scenes (dynamic point clouds or neural representations) from monocular or multi-view video inputs.
• Enable editing capabilities, such as object removal, replacement, or scene manipulation, while maintaining spatiotemporal coherence.
• Evaluate existing methods and propose improvements or adaptations for specific application scenarios.
• Enable editing capabilities, such as object removal, replacement, or scene manipulation, while maintaining spatiotemporal coherence.
• Evaluate existing methods and propose improvements or adaptations for specific application scenarios.
Applicant's Learning Objectives
• A working prototype or codebase demonstrating 4D reconstruction and editing capabilities.
• A final written report summarizing the research findings, methodology, and results .
• A research poster for presentation during UROP Week .
• An oral presentation at the UROP mini-conference .
• A final written report summarizing the research findings, methodology, and results .
• A research poster for presentation during UROP Week .
• An oral presentation at the UROP mini-conference .
Complexity of the project
Moderate