Dynamic Gaussians Mesh:
Consistent Mesh Reconstruction from Monocular Videos
ICLR 2025
Isabella Liu,
Hao Su† ,
Xiaolong Wang†
UC San Diego
† denotes equal advisory
DG-Mesh reconstructs high-fidelity, time-consistent meshes for dynamic scenes with complex non-rigid deformations. Given dynamic input and camera parameters, it recovers high-quality surfaces, appearance, and vertex motion while supporting flexible topology changes. Evaluations show it significantly outperforms baselines in dynamic mesh reconstruction and rendering.
Pipeline
Training Process
4D GS Center
Anchored GS center
Mesh
Mesh Rendering
D-NeRF Results
DG-Mesh Results
Real Results on Real Data
Nerfies: Toby-sit
Nerfies: Tail
Self-captured iPhone Dataset: Tiger
Self-captured iPhone Dataset: Starbucks
NeuralActor: D2_vlad
NeuralActor: N1_lingjie_yellowpants
Full Video
BibeTex
@article{liu2024dynamic,
title={Dynamic Gaussians Mesh: Consistent Mesh Reconstruction from Monocular Videos},
author={Liu, Isabella and Su, Hao and Wang, Xiaolong},
journal={arXiv preprint arXiv:2404.12379},
year={2024}
}