Reproduce “Self-Supervised Camera Self-Calibration from Video” 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
Jiading Fang, Igor Vasiljevic, Vitor Guizilini, Rares Ambrus, Greg Shakhnarovich, Adrien Gaidon, Matthew R.Walter
A self-supervised framework for camera self-calibration leverages unified camera models and raw video data to achieve sub-pixel reprojection error, demonstrating robust online re-calibration and state-of-the-art depth estimation on the EuRoC dataset.
Key insights
- The view-synthesis objective used in self-supervised depth and ego-motion estimation can effectively guide the learning of explicit, interpretable camera intrinsic parameters.
- Unified Camera Models offer a balance of generality and computational efficiency, enabling self-calibration for diverse camera types (pinhole, fisheye, catadioptric) in a differentiable manner.
- Learning a single set of intrinsic parameters per video sequence, rather than independently for each frame, significantly improves calibration stability and accuracy.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.