Research map / Point clouds / Shape and surface completion
Learning To Measure the Point Cloud Reconstruction Loss in a Representation Space
Tianxin Huang, Zhonggan Ding, Jiangning Zhang, Ying Tai, Zhenyu Zhang, Mingang Chen, Chengjie Wang, 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 |
|---|---|---|---|
| replace PCLoss with CALoss | AE · ShapeNet test | MCD: 0.23 -> 0.21 | Improved |
| replace PCLoss with CALoss | AE · ModelNet40 | MCD: 0.59 -> 0.58 | Improved |
| dynamically update CALoss alongside the task network rather than using a pre-trained frozen loss (ablation) | AE · ShapeNet test | MCD: 0.75 -> 0.21 | Improved |
| add the contrastive constraint Lp (ablation) | CALoss · ShapeNet | MCD: 3.55 -> 0.21 | Improved |
| add perturbation-generated positive samples (ablation) | CALoss · ShapeNet | MCD: 1.83 -> 0.21 | Improved |
| use adversarial feedback from reconstructed outputs instead of minibatch distance maximization (ablation) | CALoss · ShapeNet | MCD: 0.77 -> 0.21 | Improved |
| add δ regularization (ablation) | CALoss · ShapeNet | HD: 1.63 -> 1.53 | Improved |
| use the dynamic 1/|Nσ| coefficient for Lp (ablation) | CALoss · ShapeNet | MCD: 0.21 -> 0.21 | No clear effect |
| feed both max- and average-pooled features to the Pooling Controller rather than only max-pooled features (ablation) | CALoss · ShapeNet | MCD: 0.22 -> 0.21 | Improved |
| replace max pooling with Adaptive Pooling (ablation) | CALoss · ShapeNet | MCD: 0.34 -> 0.21 | Improved |
| replace average pooling with Adaptive Pooling (ablation) | CALoss · ShapeNet | MCD: 0.61 -> 0.21 | Improved |
Closest papers by meaning
- Learning to Train a Point Cloud Reconstruction Network without Matching (ECCV 2022 · similarity 0.88)
- GAL: Geometric Adversarial Loss for Single-View 3D-Object Reconstruction (ECCV 2018 · similarity 0.88)
- Manifold Adversarial Learning for Cross-Domain 3D Shape Representation (ECCV 2022 · similarity 0.87)
- A Point Set Generation Network for 3D Object Reconstruction From a Single Image (CVPR 2017 · similarity 0.87)
- Shape-Invariant 3D Adversarial Point Clouds (CVPR 2022 · similarity 0.87)
- APML: Adaptive Probabilistic Matching Loss for Robust 3D Point Cloud Reconstruction (NeurIPS 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.
