A Survey on 3D and 4D Reconstruction and Generation for World Modeling

PhD Qualifying Examination


Title: "A Survey on 3D and 4D Reconstruction and Generation for World Modeling"

by

Mr. Zhenxing MI


Abstract:

High-fidelity world modeling is a central problem in computer vision, robotics,
and spatial AI, requiring robust methods for both reconstructing real
environments and simulating virtual ones. This survey provides a review of
state-of-the-art methods for 3D and 4D reconstruction and generation. On the
reconstruction side, we trace the evolution from classical Structure from Motion
(SfM) and Multi-View Stereo (MVS) to neural implicit representations and
scalable feedforward architectures, highlighting the shift from per-scene
optimization to generalizable models and their extension to dynamic 4D scenes.
On the generative side, we cover methods for synthesizing diverse 3D objects,
explorable 3D scenes, and dynamic 4D environments, analyzing the developments
from static object synthesis to full 4D scene simulation. Across both lines of
work, we identify common challenges in scaling data and model capacity, modeling
long-range temporal dynamics, maintaining geometric fidelity under sparse or
noisy views, and integrating foundation diffusion models with geometry-aware
representations. We conclude by outlining these challenges as key directions for
building scalable and robust World Models for downstream applications in vision
and robotics.


Date:                   Thursday, 11 December 2025

Time:                   10:00am - 12:00noon

Venue:                  Room 2128C
                        Lift 19

Committee Members:      Dr. Dan Xu (Supervisor)
                        Dr. Qifeng Chen (Chairperson)
                        Dr. Hao Chen