Research map / Point clouds / Shape and surface completion
Mapping in a Cycle: Sinkhorn Regularized Unsupervised Learning for Point Cloud Shapes
Lei Yang, Wenxi Liu, Zhiming Cui, Nenglun Chen, Wenping Wang
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 |
|---|---|---|---|
| add Sinkhorn regularization L_S (ablation) | cycle-consistency framework · ShapeNet part (Airplane/Chair/Table), val | CC% (mean val): 41.5 -> 67.5 | Improved |
| use Sinkhorn permutation relaxation instead of identity loss L_I (ablation) | cycle-consistency framework · ShapeNet part (Airplane/Chair/Table), val | CC% (mean val): 48.7 -> 67.5 | Improved |
| add self-attention modules to PointNet++ (ablation) | cycle-consistency framework · ShapeNet part (Airplane/Chair/Table), val | CC% (mean val): 50.0 -> 67.5 | Improved |
| add normals as input (ablation) | cycle-consistency framework · ShapeNet part (Airplane/Chair/Table), val | CC% (mean val): 56.3 -> 67.5 | Improved |
| add learned pointwise features to PointNet++ part segmentation (full data) | PointNet++ · ShapeNet part segmentation (full training data) | mean IoU: 85.5% vs PointNet++ baseline | Improved |
| add learned pointwise features with 10% labeled data | PointNet++ · ShapeNet part segmentation (10% training data) | mean IoU: 79.8% mean IoU, gains in 12 out of 16 categories, up to 7% | Improved |
| compare with PRNet for partial registration | registration with learned features · ModelNet40 partial registration | rotation/translation RMSE, MAE: Chair rotation RMSE 4.93 -> 6.202 (worse); Car 15.42 -> 4.678 | Mixed |
| compare with other unsupervised methods for keypoint transfer | learned pointwise features · Keypoint transfer (mean) | % keypoints < 0.05: 40.9 -> 49.4 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- PRNet — PRNet: Self-Supervised Learning for Partial-to-Partial Registration (NeurIPS 2019) · better on 9 of 10 shared results
- FoldingNet — FoldingNet: Point Cloud Auto-Encoder via Deep Grid Deformation (CVPR 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
- Shape Self-Correction for Unsupervised Point Cloud Understanding (ICCV 2021 · similarity 0.89)
- Unsupervised Multi-Task Feature Learning on Point Clouds (ICCV 2019 · similarity 0.88)
- Learning a Structured Latent Space for Unsupervised Point Cloud Completion (CVPR 2022 · similarity 0.88)
- Mining and Transferring Feature-Geometry Coherence for Unsupervised Point Cloud Registration (NeurIPS 2024 · similarity 0.87)
- FreePoint: Unsupervised Point Cloud Instance Segmentation (CVPR 2024 · similarity 0.87)
- Unsupervised Template-assisted Point Cloud Shape Correspondence Network (CVPR 2024 · 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.
