Dense-SfM: Structure from Motion with Dense Consistent Matching

JongMin Lee, SungJoo Yoo
Seoul National University
IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) 2025
Paper arXiv Code (will be open soon)
Test image description
Our Dense-SfM pipeline leverages Gaussian Splatting to offer more accurate and dense 3D points. Thanks to dense matching, Our pipeline can also be applied to texture-less objects, further improving the quality of 3D points and pose estimation.

Method

Test image description

Our framework is designed as a three-stage pipeline (see figure above):

1. Construct an initial SfM model using a two-view dense matcher (e.g. DKM or RoMa).
2. Extend each 3D point's track by projecting it onto new images using Gaussian Splatting.
3. Perform iterative SfM refinement to enhance the accuracy of poses and point cloud.

Visualization

Test image description

Left: Qualitative Results on Texture-Poor SfM dataset.
Right: Qualitative Results on IMC 2021 dataset.

Related Links

There are excellent works that is related with ours and gave us inspirations.

Detector-Free SfM proposes a detector-free structure from motion framework that eliminates the requirement of keypoint detection and can recover poses even on challenging texture-poor scenes.

DKM and RoMa propose a significantly robust dense feature matcher.

BibTeX

@misc{lee2025densesfmstructuremotiondense,
      title={Dense-SfM: Structure from Motion with Dense Consistent Matching}, 
      author={JongMin Lee and Sungjoo Yoo},
      year={2025},
      eprint={2501.14277},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/2501.14277}, 
}