Research map / 3D vision / Face and shape
Recurrent Multi-View Alignment Network for Unsupervised Surface Registration
Wanquan Feng, Juyong Zhang, Hongrui Cai, Haofei Xu, Junhui Hou, Hujun Bao
Check 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 |
|---|---|---|---|
| Represent non-rigid deformation using recurrently added rigid transforms with learned point-wise weights | RMA-Net · Deform | CD: 2.375 -> 0.599 | Improved |
| Use recurrently added rigid transforms with learned point-wise weights | RMA-Net · Deform | EMD: 5.478 -> 0.386 | Improved |
| Use multi-view depth and mask loss instead of Chamfer loss (ablation) | RMA-Net · Deformable objects | CD: 1.652 -> 0.599 | Improved |
| Use multi-view depth and mask loss instead of EMD loss (ablation) | RMA-Net · Deformable objects | CD: 3.241 -> 0.599 | Improved |
| Add mask loss to depth loss (ablation) | RMA-Net · Deformable objects | CD: 0.701 -> 0.691 | Improved |
| Increase views from 5^2 to 7^2 (ablation) | RMA-Net · Deformable objects | CD: 0.691 -> 0.618 | Improved |
| Increase views from 7^2 to 9^2 (ablation) | RMA-Net · Deformable objects | CD: 0.618 -> 0.600 | Improved |
| Increase recurrent stages from 3 to 7 (ablation) | RMA-Net · Deformable objects | CD: 10.717 -> 0.599 | Improved |
| Use ground-truth supervision instead of unsupervised training | RMA-Net · ModelNet40 | RMSE(R): 1.287 -> 0.735 | Improved |
| Use ground-truth supervision instead of unsupervised training | RMA-Net · ModelNet40 | MAE(t): 0.007 -> 0.009 | Hurt |
| Apply RMA-Net to raw scanned surfaces with noise, outliers, and incompleteness | RMA-Net · DFAUST | CD: 0.43 -> 0.24 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- 3D-Coded — 3D-CODED: 3D Correspondences by Deep Deformation (ECCV 2018) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Deep Confidence Guided Distance for 3D Partial Shape Registration (AAAI 2022 · similarity 0.87)
- Recurrent Registration Neural Networks for Deformable Image Registration (NeurIPS 2019 · similarity 0.87)
- Unsupervised Multi-Modal Image Registration via Geometry Preserving Image-to-Image Translation (CVPR 2020 · similarity 0.86)
- DPCN++: Differentiable Phase Correlation Network for Versatile Pose Registration (TPAMI 2023 · similarity 0.86)
- Occlusion-aware Non-Rigid Point Cloud Registration via Unsupervised Neural Deformation Correntropy (ICLR 2025 · similarity 0.86)
- 3D-R2N2: A Unified Approach for Single and Multi-view 3D Object Reconstruction (ECCV 2016 · similarity 0.86)
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.
