Reproduce “DP-SfM: Dual-Pixel Structure-from-Motion without Scale Ambiguity” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Lilika Makabe, Kohei Ashida, Hiroaki Santo, Fumio Okura, Yasuyuki Matsushita
This research introduces DP-SfM, a method for automatically determining the absolute scale in multi-view 3D reconstruction by utilizing dual-pixel (DP) images. By coupling SfM-derived depths with DP defocus blur information through a thin-lens model, the system achieves absolute scene scale estimation without requiring external references, calibration targets, or extensive training data, demonstrating superior accuracy compared to learning-based monocular depth estimators.
Key insights
- The fundamental scale ambiguity in Structure-from-Motion (SfM) can be resolved by jointly considering the focus distance ambiguity present in defocus blur information derived from dual-pixel images, linked by the thin-lens model.
- Dual-pixel sensors provide a unique and robust way to estimate defocus blur sizes from uncalibrated image data, making them ideal for passively extracting the depth-related cues needed for absolute scale recovery.
- A two-stage methodology, combining an efficient linear initialization with a refinement based on intensity alignment, enhances the overall accuracy and robustness of scale estimation, mitigating errors from initial blur size estimations.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.