Research map / Point clouds / Registration and matching
Improving RGB-D Point Cloud Registration by Learning Multi-Scale Local Linear Transformation
Ziming Wang, Xiaoliang Huo, Zhenghao Chen, Jing Zhang, Lu Sheng, Dong Xu
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 RGB-D concatenation features with GAVE (LLT) in UR&R, 3D Match training | UR&R (RGB-D) · ScanNet | mean rotation error (deg): 3.8 -> 3.0 | Improved |
| replace RGB-D concatenation with GAVE, 3D Match training | UR&R (RGB-D) · ScanNet | FMR: 0.78 -> 0.87 | Improved |
| replace RGB-D concatenation with GAVE, ScanNet training | UR&R (RGB-D) · ScanNet | mean rotation error (deg): 2.6 -> 2.5 | Improved |
| add multi-scale fusion (ablation) | Baseline feature extractor · ScanNet (3D Match) | mean Chamfer distance: 6.5 -> 6.3 | Improved |
| add dilated convolutions in visual feature extractor (ablation) | MS variant · ScanNet (3D Match) | mean rotation error (deg): 3.8 -> 3.6 | Improved |
| add LLT fusion module (ablation) | MS + DC variant · ScanNet (3D Match) | mean rotation error (deg): 3.6 -> 3.0 | Improved |
| add guidance map to LLT (ablation) | LLT · ScanNet (3D Match) | mean rotation error (deg): 3.2 -> 3.0 | Improved |
| use linear transformation instead of affine transformation (ablation) | LLT · ScanNet (3D Match) | mean rotation error (deg): 3.1 -> 3.0 | Improved |
| use LLT instead of linear-transformer multi-head cross attention (ablation) | Fusion module · ScanNet (3D Match) | mean rotation error (deg): 3.8 -> 3.0 | Improved |
| increase n_grid from 2 to 3 (ablation) | LLT · ScanNet (3D Match) | rotation accuracy @5deg (%): 92.6 -> 93.4 | Improved |
| increase n_grid from 3 to 4 (ablation) | LLT · ScanNet (3D Match) | mean Chamfer distance: 5.3 -> 5.7 | Hurt |
| reduce n_group from 32 to 16 (ablation) | LLT · ScanNet (3D Match) | mean rotation error (deg): 3.1 -> 3.0 | Improved |
Closest papers by meaning
- PointMBF: A Multi-scale Bidirectional Fusion Network for Unsupervised RGB-D Point Cloud Registration (ICCV 2023 · similarity 0.92)
- Zero-shot RGB-D Point Cloud Registration with Pre-trained Large Vision Model (CVPR 2025 · similarity 0.90)
- Partial Point Cloud Registration with Multi-view 2D Image Learning (AAAI 2025 · similarity 0.90)
- ColorPCR: Color Point Cloud Registration with Multi-Stage Geometric-Color Fusion (CVPR 2024 · similarity 0.89)
- FreeReg: Image-to-Point Cloud Registration Leveraging Pretrained Diffusion Models and Monocular Depth Estimators (ICLR 2024 · similarity 0.89)
- Generative Point Cloud Registration (ICML 2025 · similarity 0.88)
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.
