Research map / Point clouds / Point cloud classification
No Pain, Big Gain: Classify Dynamic Point Cloud Sequences With Static Models by Fitting Feature-Level Space-Time Surfaces
Jia-Xing Zhong, Kaichen Zhou, Qingyong Hu, Bing Wang, Niki Trigoni, Andrew Markham
Check 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 the Kinet temporal stream with feature-level ST-surfaces to static PointNet++ | PointNet++ · SHREC'17, w/o BBox | accuracy: from only 86.5% to 94.1% | Improved |
| Add the Kinet temporal stream with feature-level ST-surfaces to PointNet++ | PointNet++ · MSRAction-3D, 24 frames | accuracy: 93.27% | Improved |
| Pretrain the static PointNet++ backbone on ModelNet40 (ablation) | Static PointNet++ spatial stream · NvGesture | accuracy: 82.6% to 84.5% | Improved |
| Pretrain the static backbone that supplies features to the temporal stream (ablation) | Kinet temporal stream · NvGesture | accuracy: 80.9 to 82.4 | Improved |
| Fuse predictions from the pretrained spatial and temporal streams (ablation) | Kinet · NvGesture | accuracy: 89.1% | Improved |
| Add ST-normal estimation to the feature abstraction operation (ablation) | Kinet · NvGesture | accuracy: 85.3% to 89.1% | Improved |
| Add learned weighted-neighbor grouping instead of fixed unit weights (ablation) | Kinet · NvGesture | accuracy: 87.9% to 89.1% | Improved |
| Use multi-frame temporal neighborhoods rather than a single-frame neighborhood (ablation) | Kinet temporal branch · NvGesture | accuracy | Improved |
| Set the temporal radius to 1 rather than other searched values (ablation) | Kinet temporal branch · NvGesture | accuracy | Improved |
| Use group-wise feature dimension d=4 (ablation) | Kinet temporal branch · NvGesture | accuracy | Mixed |
| Use a 50% feature-dimension reduction ratio (ablation) | Kinet temporal branch · NvGesture | accuracy | No clear effect |
| Use a spatial radius of 0.5 (ablation) | Kinet temporal branch · NvGesture | accuracy | No clear effect |
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 4 of 5 shared results
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
- Dynamic 3D Point Cloud Sequences as 2D Videos (TPAMI 2024 · similarity 0.89)
- FlowNet3D: Learning Scene Flow in 3D Point Clouds (CVPR 2019 · similarity 0.88)
- PointFlowNet: Learning Representations for Rigid Motion Estimation From Point Clouds (CVPR 2019 · similarity 0.88)
- IDEA-Net: Dynamic 3D Point Cloud Interpolation via Deep Embedding Alignment (CVPR 2022 · similarity 0.88)
- LinNet: Linear Network for Efficient Point Cloud Representation Learning (NeurIPS 2024 · similarity 0.86)
- 3DInAction: Understanding Human Actions in 3D Point Clouds (CVPR 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.
