Research map / Point clouds / Registration and matching
A Robust Loss for Point Cloud Registration
Zhi Deng, Yuxin Yao, Bailin Deng, Juyong Zhang
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| replace Chamfer distance with the line-intersection metric | optimization-based registration · Human (optimization-based) | rotation error (degrees): 5.863 -> 0.576 | Improved |
| add Welsch robust function to Chamfer distance (nu0=0.5) | Chamfer distance · Human (optimization-based) | rotation error (degrees): 5.863 -> 4.84 | Improved |
| use convex combination of candidate points for intersection instead of all candidates (ablation) | line-intersection metric · Human (optimization-based) | rotation error (degrees): 49.27 -> 0.576 | Improved |
| sample random lines uniformly instead of bounding-box sampling (Sample1) (ablation) | line-intersection metric · Human (optimization-based) | rotation error (degrees): 20.482 -> 0.576 | Improved |
| sample random lines uniformly instead of perturbed point sampling (Sample2) (ablation) | line-intersection metric · Human (optimization-based) | rotation error (degrees): 5.201 -> 0.576 | Improved |
| set nu0 to 0.5 instead of 1 (ablation) | line-intersection metric · Human (optimization-based) | rotation error (degrees): 2.814 -> 0.576 | Improved |
| set nu0 to 0.5 instead of 0.01 or 100 (ablation) | line-intersection metric · Human (optimization-based) | rotation error (degrees): 10.814 / 7.786 -> 0.576 | Improved |
| use our metric instead of Chamfer distance as unsupervised loss | DCP · Airplane | rotation error (degrees): 6.612 -> 3.808 | Improved |
| use our metric instead of Chamfer distance as unsupervised loss | FMR · Real dataset (unlabeled) | rotation error (degrees): 7.559 -> 3.263 | Improved |
| use our metric instead of Chamfer distance as unsupervised loss | RPM-Net · Real dataset (unlabeled) | rotation error (degrees): 11.28 -> 2.972 | Improved |
| unsupervised our metric instead of ground-truth supervision | RPM-Net · Airplane | rotation error (degrees): 2.19 -> 1.673 | Improved |
| unsupervised our metric instead of ground-truth supervision | DCP · Human (learning-based) | rotation error (degrees): 3.841 -> 4.841 | Hurt |
Closest papers by meaning
- Feature-Metric Registration: A Fast Semi-Supervised Approach for Robust Point Cloud Registration Without Correspondences (CVPR 2020 · similarity 0.92)
- SDRSAC: Semidefinite-Based Randomized Approach for Robust Point Cloud Registration Without Correspondences (CVPR 2019 · similarity 0.90)
- Geometric Transformer for Fast and Robust Point Cloud Registration (CVPR 2022 · similarity 0.90)
- Reliable Inlier Evaluation for Unsupervised Point Cloud Registration (AAAI 2022 · similarity 0.89)
- A robust inlier identification algorithm for point cloud registration via $\mathbf{\ell_0}$-minimization (NeurIPS 2024 · similarity 0.88)
- Inlier Confidence Calibration for Point Cloud Registration (CVPR 2024 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
