Learning Dynamic 3D Gaussians from Monocular Videos without Camera Poses

Tianrui Zhang, Yang Fu, Xiaolong Wang

Keywords: Dynamic reconstruction, camera pose estimation

Abstract: Dynamic scene reconstruction aims to recover the time-varying geometry and appearance of a dynamic scene. Existing methods, however, heavily rely on the existence of multiple-view captures or the accurate camera poses estimated by Structure from Motion (SfM) algorithms. To relax this constraint, we introduce a method capable of reconstructing generic dynamic scenes, from casually captured monocular videos without known camera poses. Unlike recent works that treat static and dynamic content separately, we propose a unified Hexplane-based Gaussian field to capture the complex effects of scene deformation and camera motion. The Hexplane decomposition enables feasible disentanglement for effective optimization. Combined with an efficient camera pose initialization strategy, our approach significantly improves view synthesis quality and camera pose estimation accuracy over previous methods, while enhancing computational efficiency.

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: 2017