Research map / Point clouds / Point cloud classification
Masked Spatio-Temporal Structure Prediction for Self-supervised Learning on Point Cloud Videos
Zhiqiang Shen, Xiaoxiao Sheng, Hehe Fan, Longguang Wang, Yulan Guo, Qiong Liu, Hao Wen, 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 |
|---|---|---|---|
| add MaST-Pre pre-training then fine-tune | P4Transformer · NTU-RGBD cross-subject | accuracy: 90.2 -> 90.8 | Improved |
| add MaST-Pre pre-training then fine-tune | P4Transformer · MSRAction-3D | accuracy: 90.94 -> 91.29 | Improved |
| add MaST-Pre pre-training then fine-tune | PST-Transformer · MSRAction-3D | accuracy: 93.73 -> 94.08 | Improved |
| transfer NTU-RGBD pre-trained encoder to gesture recognition (30 epochs) | P4Transformer · NvGesture | accuracy: 84.8 -> 87.6 | Improved |
| transfer NTU-RGBD pre-trained encoder to gesture recognition (30 epochs) | P4Transformer · SHREC'17 | accuracy: 87.5 -> 90.2 | Improved |
| add motion (temporal cardinality difference) prediction stream (ablation) | MaST-Pre · 10% NTU-RGBD | accuracy: 70.13 -> 78.25 | Improved |
| add appearance stream only pre-training (ablation) | P4Transformer · 10% NTU-RGBD | accuracy: 65.85 -> 70.13 | Improved |
| use 75% masking ratio instead of 65% or 85% (ablation) | MaST-Pre · 10% NTU-RGBD | accuracy: 76.91 / 77.20 -> 78.25 | Improved |
| use decoupled instead of coupled reconstruction loss in point tubes (ablation) | appearance stream · 10% NTU-RGBD | accuracy: 58.60 -> 78.25 | Improved |
| use 8 sections instead of 16 in cardinality difference (ablation) | MaST-Pre · 10% NTU-RGBD | accuracy: 69.50 -> 78.25 | Improved |
| add interpolation in cardinality difference (ablation) | MaST-Pre · 10% NTU-RGBD | accuracy: 76.69 -> 78.25 | Improved |
| use temporal stride 1 instead of 2 (ablation) | cardinality difference · 10% NTU-RGBD | accuracy: 75.85 -> 78.25 | Improved |
Closest papers by meaning
- Masked Motion Prediction with Semantic Contrast for Point Cloud Sequence Learning (ECCV 2024 · similarity 0.90)
- PointCMP: Contrastive Mask Prediction for Self-Supervised Learning on Point Cloud Videos (CVPR 2023 · similarity 0.89)
- Masked Discrimination for Self-Supervised Learning on Point Clouds (ECCV 2022 · similarity 0.88)
- GeoMAE: Masked Geometric Target Prediction for Self-Supervised Point Cloud Pre-Training (CVPR 2023 · similarity 0.88)
- Adapting Pre-trained 3D Models for Point Cloud Video Understanding via Cross-frame Spatio-temporal Perception (CVPR 2025 · similarity 0.87)
- 3D Feature Prediction for Masked-AutoEncoder-Based Point Cloud Pretraining (ICLR 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.
