Abstract:

We propose a novel framework for scene decomposition and static background reconstruction from everyday videos. By integrating the trained motion masks and modeling the static scene as Gaussian splats with dynamics-aware optimization, our method achieves more accurate background reconstruction results than previous works. Our proposed method is termed DAS3R, an abbreviation for Dynamics-Aware Gaussian Splatting for Static Scene Reconstruction. Compared to existing methods, DAS3R is more robust in complex motion scenarios, capable of handling videos where dynamic objects occupy a significant portion of the scene, and does not require camera pose inputs or point cloud data from SLAM-based methods.

Novel View Testing Results:

DAVIS: Testing Frames

(Click left or right to view all test image comparisons.)

DAVIS: Comparisons on PSNR

Sintel: Testing Frames

(Click left or right to view all test image comparisons.)

Sintel: Comparisons on PSNR

Dynamic Mask Prediction:

DAVIS: Dynamic Masks

(Click left or right to view all test image comparisons.)

Sintel: Dynamic Masks

(Click left or right to view all test image comparisons.)

BibTex:

            
        @article{xu2024das3r,
            title={DAS3R: Dynamics-Aware Gaussian Splatting for Static Scene Reconstruction},
            author={Kai Xu and Tze Ho Elden Tse and Jizong Peng and Angela Yao},
            journal={arXiv preprint arxiv:2412.19584},
            year={2024}}