Research map / Point clouds / Shape and surface completion
CP3: Unifying Point Cloud Completion by Pretrain-Prompt-Predict Paradigm
Mingye Xu, Yali Wang, Yihao Liu, Tong He, Yu Qiao
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 IOI pretraining to generation network (ablation) | VRCNet generation · MVP (CD) | CD: 5.96 -> 5.69 | Improved |
| add SCR refinement (predict stage) after generation (ablation) | VRCNet generation · MVP (CD) | CD: 5.96 -> 5.80 | Improved |
| add SCR refinement on top of IOI-pretrained generation (ablation) | CP3 generation · MVP (CD) | CD: 5.69 -> 5.10 | Improved |
| apply IOI pretraining to refinement network too (ablation) | CP3 · MVP (CD) | CD: 5.10 -> 5.30 | Hurt |
| add semantic modulation to refinement (ablation) | refinement network · MVP (CD) | CD: 5.60 -> 5.17 | Improved |
| add multi-scale point deconvolution (ablation) | SCR network · MVP (CD) | CD: 5.17 -> 5.10 | Improved |
| use wrong random category guidance at test (ablation) | CP3 · MVP (CD) | CD: 5.10 -> 5.68 | Hurt |
| use feature filtering rather than concatenating category (ablation) | SCR modulation · MVP (CD) | CD: 5.19 -> 5.10 | Improved |
| add IOI pretraining to PCN generation network (ablation) | PCN · MVP (CD) | CD: 9.77 -> 8.51 | Improved |
| add IOI pretraining to ECG generation network (ablation) | ECG · MVP (CD) | CD: 6.64 -> 6.34 | Improved |
| use IOI as augmentation instead of pretraining (ablation) | VRCNet · MVP (CD) | CD: 5.96 -> 5.82 | Improved |
| use mirrored incompletion input for pretraining (ablation) | VRCNet pretraining · MVP (CD) | CD: 5.96 -> 7.59 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 3 of 3 on most shared results.
- SnowFlakeNet — SnowflakeNet: Point Cloud Completion by Snowflake Point Deconvolution With Skip-Transformer (ICCV 2021) · better on 4 of 4 shared results
- PoinTr — PoinTr: Diverse Point Cloud Completion With Geometry-Aware Transformers (ICCV 2021) · better on 2 of 2 shared results
- GRNet — GRNet: Gridding Residual Network for Dense Point Cloud Completion (ECCV 2020) · better on 2 of 2 shared results
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
- A Conditional Point Diffusion-Refinement Paradigm for 3D Point Cloud Completion (ICLR 2022 · similarity 0.89)
- Cascaded Refinement Network for Point Cloud Completion with Self-supervision (TPAMI 2021 · similarity 0.89)
- Complete Structure Guided Point Cloud Completion via Cluster- and Instance-Level Contrastive Learning (NeurIPS 2025 · similarity 0.89)
- Semantic Guided Part Relation-aware Network for Point Cloud Completion (AAAI 2026 · similarity 0.89)
- Generated and Pseudo Content guided Prototype Refinement for Few-shot Point Cloud Segmentation (NeurIPS 2024 · similarity 0.88)
- Exploring Vision Semantic Prompt for Efficient Point Cloud Understanding (ICML 2025 · 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.
