Reproduce “Deep Learning-Based Point Cloud Registration: A Comprehensive Survey and Taxonomy” 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
Yu-Xin Zhang, Jie Gui, Baosheng Yu, Xiaofeng Cong, Xin Gong, Wenbing Tao, Dacheng Tao
This paper offers a comprehensive survey and a novel taxonomy for over 100 deep learning-based point cloud registration methods, ranging from supervised to unsupervised approaches. It also quantitatively evaluates state-of-the-art algorithms on benchmark datasets like 3DMatch and KITTI under unified settings, revealing performance trade-offs and future research avenues.
Key insights
- Supervised DL-PCR methods currently dominate, driven by accuracy, while unsupervised approaches are gaining interest to reduce reliance on labeled data.
- DL-PCR models frequently specialize in specific pipeline stages or integrate traditional geometric algorithms to enhance robustness and efficiency.
- Quantitative evaluation reveals trade-offs between high registration recall, inlier ratio, and model complexity across different DL-PCR algorithms.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.