ShapeGaussian: Dynamic Gaussian Splatting for Monocular Videos with Non-Parametric Shape Regularization

Zhenxiao Liang, Ning Zhang, Youbao Tang, Ruei-Sung Lin, Qixing Huang, Mei Han, Jing Xiao

Keywords: scene reconstruction, Gaussian splatting, monocular video, shape prior, 2d keypoints

TL;DR: We present a novel method for dynamic scene reconstruction with non-parametric shape prior

Abstract: In this paper, we tackle the challenging and underconstrained problem of reconstructing dynamic objects from monocular videos using a new method we term ShapeGaussian. This approach incorporates shape priors to enhance reconstruction accuracy and chance of success with few multi-view clues. Our methodology employs a two-phase process. In the first stage, we establish a temporally consistent deformation model across frames based on depth maps derived from a pre-trained estimator. The second stage obtains high-quality photo-realistic reconstruction by optimizing 3D Gaussian jointly with non-parametric shape models. Through rendering this combined model into radiance fields, we achieve high-quality, photo-realistic reconstructions of dynamically deforming objects that maintain 3D consistency across novel views. Our results demonstrate that significant improvement over previous methods on human dynamics, particularly in scenarios with scarce multi-view cues, highlighting the persistent challenges and varied approaches in recent research aimed at this inherently complex task.

Primary Area: applications to computer vision, audio, language, and other modalities

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.

Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.

Reciprocal Reviewing: I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Submission Number: 11427