Research map / Point clouds / Point cloud classification
PointChain: Learning Generalizable Point Cloud Representations via Structural Chain Modeling
Luyao Wang, Chuxin Wang, Qiao Li, Tianzhu Zhang
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 |
|---|---|---|---|
| use PointNet++-like token embedding (ablation) | PointChain (Mamba) · ScanObjectNN PB-T50-RS | accuracy (%): 90.11 -> 90.53 | Improved |
| add structural chain and next-scale prediction pretraining (ablation) | PointChain (Mamba) · ScanObjectNN PB-T50-RS | accuracy (%): 90.53 -> 92.96 | Improved |
| add cross-scale semantic alignment contrastive task (ablation) | PointChain (Mamba) · ScanObjectNN PB-T50-RS | accuracy (%): 92.96 -> 93.75 | Improved |
| add multi-scale token embedding (ablation) | PointChain (Mamba) · ScanObjectNN PB-T50-RS | accuracy (%): 92.26 -> 92.75 | Improved |
| add scale identifier (ablation) | PointChain (Mamba) · ScanObjectNN PB-T50-RS | accuracy (%): 92.75 -> 93.34 | Improved |
| add dynamic scaling strategy (ablation) | PointChain (Mamba) · ScanObjectNN PB-T50-RS | accuracy (%): 93.34 -> 93.75 | Improved |
| use next-scale prediction instead of masked point modeling (ablation) | PointChain (Mamba) · ScanObjectNN PB-T50-RS | accuracy (%): 92.09 -> 92.96 | Improved |
| set fine-tuning regularization weight lambda to 2 instead of 0 (ablation) | PointChain (Mamba) · ScanObjectNN PB-T50-RS | accuracy (%): 91.74 -> 93.75 | Improved |
| set lambda to 2 instead of 4 (ablation) | PointChain (Mamba) · ScanObjectNN PB-T50-RS | accuracy (%): 93.44 -> 93.75 | Improved |
| use representative points [1 32 128] instead of [1 16 64] (ablation) | PointChain (Mamba) · ScanObjectNN OBJ-BG | accuracy (%): 94.84 -> 95.70 | Improved |
| use representative points [1 64 256] instead of [1 16 64] (ablation) | PointChain (Mamba) · ScanObjectNN PB-T50-RS | accuracy (%): 93.75 -> 93.03 | Hurt |
| swap Transformer for Mamba backbone (ablation) | PointChain (Mamba) · ScanObjectNN PB-T50-RS | accuracy (%): 91.50 -> 93.75 | Improved |
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.
- PointMamba — PointMamba: A Simple State Space Model for Point Cloud Analysis (NeurIPS 2024) · better on 5 of 5 shared results
- PointM2AE — Point-M2AE: Multi-scale Masked Autoencoders for Hierarchical Point Cloud Pre-training (NeurIPS 2022) · better on 2 of 2 shared results
- PointBERT — Point-BERT: Pre-Training 3D Point Cloud Transformers With Masked Point Modeling (CVPR 2022) · 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
- StruMamba3D: Exploring Structural Mamba for Self-supervised Point Cloud Representation Learning (ICCV 2025 · similarity 0.88)
- PointNeXt: Revisiting PointNet++ with Improved Training and Scaling Strategies (NeurIPS 2022 · similarity 0.88)
- Point Ladder Tuning: Parameter-Efficient Hierarchical Adaptation for 3D Point Cloud Understanding (ECCV 2026 · similarity 0.87)
- Sonata: Self-Supervised Learning of Reliable Point Representations (CVPR 2025 · similarity 0.87)
- Beyond Downstream Scores: Controlled Diagnostics for Point-Cloud Self-Supervised Learning Evaluation (NeurIPS 2026 · similarity 0.87)
- Global-Local Bidirectional Reasoning for Unsupervised Representation Learning of 3D Point Clouds (CVPR 2020 · 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.
