🔧 Depth Watertight Mesh
Novel geometric representation that models both visible and occluded regions for consistent extreme viewpoint synthesis.
Generating high-quality camera-controllable videos from monocular input is challenging, particularly under extreme viewpoints. Existing methods often struggle with geometric inconsistencies and occlusion artifacts. We introduce EX-4D, a novel framework that addresses these challenges through a Depth Watertight Mesh representation that explicitly models both visible and occluded regions, ensuring geometric consistency even in extreme camera poses.
Our approach includes a simulated masking strategy that generates effective training data from monocular videos, eliminating the need for paired multi-view datasets. A lightweight LoRA-based video diffusion adapter synthesizes high-quality, physically consistent, and temporally coherent videos.
Demonstration of EX-4D's extreme viewpoint synthesis capabilities showing smooth camera transitions and consistent temporal dynamics across challenging viewpoint changes.
Novel geometric representation that models both visible and occluded regions for consistent extreme viewpoint synthesis.
Training strategy that creates effective data from monocular videos without requiring multi-view datasets.
LoRA-based video diffusion adapter with only 1% trainable parameters for efficient high-quality synthesis.
Our EX-4D framework transforms monocular videos into extreme viewpoint 4D videos through three key steps: (1) constructing a DW-Mesh as geometric prior to handle boundary occlusions, (2) generating training masks to simulate novel view occlusions, and (3) using a lightweight video diffusion adapter for physically consistent and temporally coherent synthesis.
@misc{hu2025ex4dextremeviewpoint4d,
title={EX-4D: EXtreme Viewpoint 4D Video Synthesis via Depth Watertight Mesh},
author={Tao Hu and Haoyang Peng and Xiao Liu and Yuewen Ma},
year={2025},
eprint={2506.05554},
archivePrefix={arXiv},
primaryClass={cs.CV},
url={https://arxiv.org/abs/2506.05554},
}