Abstract
COLMAP-free 3D Gaussian Splatting (3D-GS) has recently attracted increasingattention due to its remarkable performance in reconstructing high-quality 3Dscenes from unposed images or videos. However, it often struggles to handlescenes with complex camera trajectories as featured by drastic rotation andtranslation across adjacent camera views, leading to degraded estimation ofcamera poses and further local minima in joint optimization of camera poses and3D-GS. We propose PCR-GS, an innovative COLMAP-free 3DGS technique thatachieves superior 3D scene modeling and camera pose estimation via camera poseco-regularization. PCR-GS achieves regularization from two perspectives. Thefirst is feature reprojection regularization which extracts view-robust DINOfeatures from adjacent camera views and aligns their semantic information forcamera pose regularization. The second is wavelet-based frequencyregularization which exploits discrepancy in high-frequency details to furtheroptimize the rotation matrix in camera poses. Extensive experiments overmultiple real-world scenes show that the proposed PCR-GS achieves superiorpose-free 3D-GS scene modeling under dramatic changes of camera trajectories.