Research map / Point clouds / Shape and surface completion
Learning to Train a Point Cloud Reconstruction Network without Matching
Tianxin Huang, Xuemeng Yang, Jiangning Zhang, Jinhao Cui, Hao Zou, Jun Chen, Xiangrui Zhao, Yong Liu
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 log term to the PCLossNet loss (ablation) | LNSA · ShapeNet | HD: 1.91 -> 1.90 | Improved |
| add uniformity constraint L_UC (ablation) | LNSA · ShapeNet | HD: 1.90 -> 1.72 | Improved |
| add decay radius constraint L_R (ablation) | LNSA · ModelNet40 | HD: 5.52 -> 5.30 | Improved |
| add decay radius constraint L_R (ablation) | LNSA · ShapeNet | MCD: 0.23 -> 0.23 | No clear effect |
| replace matching loss CD with LNSA | AE · ShapeNet | MCD: 0.32 -> 0.23 | Improved |
| replace EMD with LNSA | AE · ShapeNet | HD: 2.23 -> 1.66 | Improved |
| use sampling head ACSA instead of fully-connected head ACFC (ablation) | PCLossNet on Local AE · ShapeNet | MCD: 0.15 -> 0.13 | Improved |
| replace CD with LNSA for unsupervised features | AE · ModelNet10 SVM classification | accuracy: 90.60 -> 91.48 | Improved |
| replace CD with LNFC | Folding · ModelNet10 SVM classification | accuracy: 91.03 -> 89.93 | Hurt |
| replace discriminator-plus-matching losses (CRN) with LNSA | AE · ShapeNet | MCD: 0.31 -> 0.23 | Improved |
| replace EMD with LNSA for training speed | AE · ShapeNet training time | time per iteration (ms): 216 -> 57 | Improved |
| use single random backward pass per iteration (FLNSA) (ablation) | LNSA · ShapeNet | time and MCD: 57 -> 23 ms, MCD 0.23 -> 0.24 | Mixed |
Closest papers by meaning
- Learning To Measure the Point Cloud Reconstruction Loss in a Representation Space (CVPR 2023 · similarity 0.88)
- CAPNet: Continuous Approximation Projection for 3D Point Cloud Reconstruction Using 2D Supervision (AAAI 2019 · similarity 0.87)
- A Point Set Generation Network for 3D Object Reconstruction From a Single Image (CVPR 2017 · similarity 0.87)
- DC-PCN: Point Cloud Completion Network with Dual-Codebook Guided Quantization (AAAI 2025 · similarity 0.86)
- APML: Adaptive Probabilistic Matching Loss for Robust 3D Point Cloud Reconstruction (NeurIPS 2025 · similarity 0.86)
- A Computation-Aware Shape Loss Function for Point Cloud Completion (AAAI 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.
