CasualHDR: Robust High Dynamic Range 3D Gaussian Splatting from Casually Captured Videos
Keywords: High Dynamic Range, 3D Reconstruction, Continous-time Trajectory, Radiance Field, Motion Blur
Abstract: In recent years, thanks to innovations in 3D scene representation, novel view synthesis and photo-realistic dense 3D reconstruction from multi-view images, such as neural radiance field (NeRF) and 3D Gaussian Splatting (3DGS), have garnered widespread attention due to their superior performance. However, most works rely on low dynamic range (LDR) images and representations of scenes, which limits the capturing of richer scene details. Prior works have focused on high dynamic range (HDR) scene recovery, typically require repeatedly capturing of multiple sharp images with different exposure times at fixed camera positions, which is time-consuming and challenging in practice.For a more flexible data acquisition, we propose a one-stage method: \textbf{CasualHDR} to easily and robustly recover the 3D HDR scene from casual videos with auto-exposure (AE) enabled, even in the presence of severe motion blur and varying exposure time. CasualHDR contains a unified differentiable physical imaging model which jointly optimize (i.e. bundle adjust) exposure time, camera response function (CRF), continuous-time camera motion trajectory on $\mathbb{SE}(3)$, and the 3DGS-based HDR scene. Extensive experiments demonstrate that our approach outperforms existing reconstruction methods in terms of robustness and rendering quality. Three applications can be achieved after the 3DGS HDR scene reconstruction: novel-view synthesis, image deblurring (deblur input images) and HDR editing (adjust the exposure time thus brightness of the input images).
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: 6790
Loading