Research map / Point clouds / Point cloud classification
PointCMP: Contrastive Mask Prediction for Self-Supervised Learning on Point Cloud Videos
Zhiqiang Shen, Xiaoxiao Sheng, Longguang Wang, Yulan Guo, Qiong Liu, Xi Zhou
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 |
|---|---|---|---|
| Pre-train with PointCMP, then fine-tune PSTNet | PSTNet · MSRAction-3D, 8 frames | accuracy (%): 83.50% to 89.56% | Improved |
| Pre-train with PointCMP, then use a frozen encoder with a linear classifier | PSTNet · MSRAction-3D, 8 and 12 frames | accuracy (%): 83.50% to 88.55%; 87.88% to 90.24% | Improved |
| Use PointCMP pre-training with 50% of NTU-RGBD training data | PSTNet · NTU-RGBD cross-subject | accuracy (%): 88.5% vs. 90.5% | Mixed |
| Transfer PointCMP-pretrained representations to PSTNet | PSTNet · NvGesture, 35 epochs | accuracy (%): 78.9% vs. 84.0% | Improved |
| Use both local and global branches rather than either branch alone (ablation) | PointCMP · MSRAction-3D | accuracy (%): 89.22% and 49.49% -> 89.76% | Improved |
| Add similarity-based augmentation to the local and global branches (ablation) | PointCMP · MSRAction-3D | accuracy (%): 89.76% -> 91.92% | Improved |
| Use segment-wise rather than token-wise masking (ablation) | PointCMP · MSRAction-3D | accuracy (%) | Improved |
| Use similarity-based rather than random segment masking (ablation) | PointCMP · MSRAction-3D, segment masking at 25% and 50% | accuracy (%): 90.81%/88.15% -> 91.92%/90.24% | Improved |
| Add hard negative samples (ablation) | PointCMP global branch · MSRAction-3D | accuracy (%): 90.52% -> 91.92% | Improved |
| Generate hard negatives by principal-channel erasing rather than random channel erasing (ablation) | PointCMP global branch · MSRAction-3D | accuracy (%): 91.29 -> 91.92 | Improved |
| Add the spatio-temporal matching module (ablation) | PointCMP local branch · MSRAction-3D, local branch only | accuracy (%): 86.20 -> 89.22 | Improved |
| Add the spatio-temporal matching module (ablation) | PointCMP local and global branches · MSRAction-3D | accuracy (%): 90.24 -> 91.92 | Improved |
Later papers that compared with this method
4 later accepted papers report PointCMP as a baseline in their results tables; 3 of them report a better number on most of the results they share with it.
- Tri-Efficient Transfer Learning for Point Cloud Videos (ECCV 2026 · better on 2 of 2 shared results)
- Adapting Pre-trained 3D Models for Point Cloud Video Understanding via Cross-frame Spatio-temporal Perception (CVPR 2025 · better on 4 of 4 shared results)
- PvNeXt: Rethinking Network Design and Temporal Motion for Point Cloud Video Recognition (ICLR 2025 · better on 1 of 2 shared results)
- UST-SSM: Unified Spatio-Temporal State Space Models for Point Cloud Video Modeling (ICCV 2025 · better on 1 of 1 shared result)
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- PSTNet++ — PSTNet: Point Spatio-Temporal Convolution on Point Cloud Sequences (ICLR 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 Contrastive Prediction with Semantic Clustering for Self-Supervised Learning on Point Cloud Videos (ICCV 2023 · similarity 0.90)
- Masked Motion Prediction with Semantic Contrast for Point Cloud Sequence Learning (ECCV 2024 · similarity 0.90)
- Masked Spatio-Temporal Structure Prediction for Self-supervised Learning on Point Cloud Videos (ICCV 2023 · similarity 0.89)
- Masked Discrimination for Self-Supervised Learning on Point Clouds (ECCV 2022 · similarity 0.89)
- 3D Feature Prediction for Masked-AutoEncoder-Based Point Cloud Pretraining (ICLR 2024 · similarity 0.85)
- Asymmetric Dual Self-Distillation for 3D Self-Supervised Representation Learning (NeurIPS 2025 · similarity 0.85)
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.
