Research map / Point clouds / Shape and surface completion
DiffPMAE: Diffusion Masked Autoencoders for Point Cloud Reconstruction
Yanlong LI, Chamara Madarasingha, Kanchana Thilakarathna
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 |
|---|---|---|---|
| train loss on entire object (masked + visible) instead of masked regions only (ablation) | DiffPMAE · ShapeNet-55, r=75% | MMD CD (x1e-3): 1.947 -> 1.125 (entire-object eval); masked-only eval is worse for setting (a) | Mixed |
| increase mask ratio from 40% to 75% (ablation) | DiffPMAE · ShapeNet-55 masked-only | MMD CD (x1e-3): 6.187 -> 2.847 | Improved |
| use block masking instead of random masking (ablation) | DiffPMAE · ShapeNet-55, r=75% | MMD CD (x1e-3): 1.125 -> 1.051 | No clear effect |
| use G64 N32 patching instead of G32 N64 or G128 N16 (ablation) | DiffPMAE · ShapeNet-55 | MMD CD (x1e-3): 1.441 -> 1.125 (vs G32 N64); 1.604 -> 1.125 (vs G128 N16) | Improved |
| increase latent width from 192 to 384 (ablation) | DiffPMAE · ShapeNet-55 | MMD CD (x1e-3): 1.370 -> 1.125 | Improved |
| increase latent width from 384 to 768 (ablation) | DiffPMAE · ShapeNet-55 | MMD CD (x1e-3): 1.125 -> 1.043 | No clear effect |
| increase diffusion steps from 200 to 300 (ablation) | DiffPMAE · ShapeNet-55 | MMD CD (x1e-3): 1.125 -> 1.506 | Hurt |
| reduce diffusion steps from 200 to 100 (ablation) | DiffPMAE · ShapeNet-55 | MMD CD (x1e-3): 1.125 -> 1.449 | Hurt |
| train longer (600 epochs), larger decoder at t=300 (ablation) | DiffPMAE · ShapeNet-55 | MMD CD (x1e-3): 1.506 -> 2.128 (e=600); 1.792 (d=8,h=6,e=600) | No clear effect |
| add MAE position embedding as conditional input in completion (ablation) | DiffPMAE · Completion, mask 75% | MMD CD (x1e-3): 4.403 -> 1.901 | Improved |
| pre-train on ShapeNet instead of training from scratch on ScanObjectNN (ablation) | DiffPMAE · ScanObjectNN with background | MMD CD (x1e-3): 1.685 -> 1.120 | Improved |
| fine-tune ShapeNet-pretrained model on ScanObjectNN (ablation) | DiffPMAE · ScanObjectNN with background | MMD CD (x1e-3): 1.120 -> 1.244 | No clear effect |
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.
- PointM2AE — Point-M2AE: Multi-scale Masked Autoencoders for Hierarchical Point Cloud Pre-training (NeurIPS 2022) · better on 2 of 3 shared results
- PoinTr — PoinTr: Diverse Point Cloud Completion With Geometry-Aware Transformers (ICCV 2021) · 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
- Point-MaDi: Masked Autoencoding with Diffusion for Point Cloud Pre-training (NeurIPS 2025 · similarity 0.91)
- 3D Feature Prediction for Masked-AutoEncoder-Based Point Cloud Pretraining (ICLR 2024 · similarity 0.89)
- RI-MAE: Rotation-Invariant Masked AutoEncoders for Self-Supervised Point Cloud Representation Learning (AAAI 2025 · similarity 0.89)
- SPU-IMR: Self-supervised Arbitrary-scale Point Cloud Upsampling via Iterative Mask-recovery Network (AAAI 2025 · similarity 0.88)
- Adaptive Siamese Masked Autoencoder with Global Optimization for Unsupervised Point Cloud Shape Correspondence (AAAI 2025 · similarity 0.88)
- PointInfinity: Resolution-Invariant Point Diffusion Models (CVPR 2024 · similarity 0.88)
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.
