Research map / Point clouds / Shape and surface completion
PoinTr: Diverse Point Cloud Completion With Geometry-Aware Transformers
Xumin Yu, Yongming Rao, Ziyi Wang, Zuyan Liu, Jiwen Lu, Jie Zhou
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 query generator between encoder and decoder (ablation) | PoinTr (vanilla transformer) · PCN dataset | CD-l1 x1000: 9.43 -> 9.09 | Improved |
| add DGCNN feature extractor (ablation) | PoinTr with query generator · PCN dataset | CD-l1 x1000: 9.09 -> 8.69 | Improved |
| add geometry-aware block to all transformer blocks (ablation) | PoinTr with query generator and DGCNN · PCN dataset | CD-l1 x1000: 8.69 -> 8.44 | Improved |
| restrict geometry-aware block to the first transformer block only (instead of all) (ablation) | PoinTr · PCN dataset | CD-l1 x1000: 8.44 -> 8.38 | Improved |
| add query generator (ablation) | PoinTr (vanilla transformer) · PCN dataset | F-Score@1%: 67.82 -> 0.713 | Improved |
| add DGCNN feature extractor (ablation) | PoinTr with query generator · PCN dataset | F-Score@1%: 0.713 -> 0.736 | Improved |
| add geometry-aware block to first block only (ablation) | PoinTr with query generator and DGCNN · PCN dataset | F-Score@1%: 0.736 -> 0.745 | Improved |
| use PoinTr instead of GRNet | completion network · ShapeNet-55 | CD-l2 avg x1000: 1.97 -> 1.09 | Improved |
| use PoinTr instead of GRNet | completion network · ShapeNet-55 | F-Score@1%: 0.238 -> 0.464 | Improved |
| evaluate on unseen instead of seen categories | PoinTr · ShapeNet-34 | CD-l2 avg x1000: 1.23 -> 2.05 | Hurt |
| use PoinTr instead of CRN | completion network · PCN dataset | CD-l1 x1000: 8.51 -> 8.38 | Improved |
| use PoinTr instead of GRNet | completion network · KITTI LiDAR cars | Fidelity: 0.816 -> 0.000 | Improved |
Later papers that compared with this method
27 later accepted papers report PoinTr as a baseline in their results tables; 23 of them report a better number on most of the results they share with it.
- DANCE: Density-agnostic and Class-aware Network for Point Cloud Completion (AAAI 2026 · better on 7 of 7 shared results)
- UBone3D: Physics-Rectified Conditional Flow Matching for Anatomical 3D Shape Completion from Ultrasound (ECCV 2026 · better on 5 of 5 shared results)
- BuildAnyPoint: 3D Building Structured Abstraction from Diverse Point Clouds (CVPR 2026 · better on 4 of 4 shared results)
- Rethinking Multimodal Point Cloud Completion: A Completion-by-Correction Perspective (AAAI 2026 · better on 2 of 2 shared results)
- PQDT: Pseudo-Query Dual Transformer for Robust Point Cloud Restoration (CVPR 2026 · better on 1 of 1 shared result)
- DC-PCN: Point Cloud Completion Network with Dual-Codebook Guided Quantization (AAAI 2025 · better on 1 of 1 shared result)
- ESCAPE: Equivariant Shape Completion via Anchor Point Encoding (CVPR 2025 · better on 1 of 1 shared result)
- EINet: Point Cloud Completion via Extrapolation and Interpolation (ECCV 2024 · better on 2 of 2 shared results)
19 more not listed.
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 4 of 4 on most shared results.
- GRNet — GRNet: Gridding Residual Network for Dense Point Cloud Completion (ECCV 2020) · better on 12 of 12 shared results
- PMP-Net — PMP-Net: Point Cloud Completion by Learning Multi-Step Point Moving Paths (CVPR 2021) · better on 1 of 1 shared result
- PFNet — PF-Net: Point Fractal Network for 3D Point Cloud Completion (CVPR 2020) · better on 1 of 1 shared result
- 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
- AdaPoinTr: Diverse Point Cloud Completion With Adaptive Geometry-Aware Transformers (TPAMI 2023 · similarity 0.97)
- PointCFormer: A Relation-Based Progressive Feature Extraction Network for Point Cloud Completion (AAAI 2025 · similarity 0.90)
- SymmCompletion: High-Fidelity and High-Consistency Point Cloud Completion with Symmetry Guidance (AAAI 2025 · similarity 0.90)
- Position-Aware Guided Point Cloud Completion with CLIP Model (AAAI 2025 · similarity 0.89)
- Point Transformer V3: Simpler Faster Stronger (CVPR 2024 · similarity 0.89)
- DAPoinTr: Domain Adaptive Point Transformer for Point Cloud Completion (AAAI 2025 · similarity 0.89)
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.
