Research map / Point clouds / Point cloud classification
Human Interaction Learning on 3D Skeleton Point Clouds for Video Violence Recognition
Yukun Su, Guosheng Lin, Jinhui Zhu, Qingyao Wu
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 SPIL interaction learning with Euclidean-distance position term (ablation) | point-convolution baseline · RWF-2000 | accuracy: 84.3 -> 86.4 | Improved |
| add SPIL with distance spanning position term (ablation) | point-convolution baseline · RWF-2000 | accuracy: 84.3 -> 88.0 | Improved |
| add SPIL with distance masking position term (ablation) | point-convolution baseline · RWF-2000 | accuracy: 84.3 -> 88.7 | Improved |
| use masking instead of spanning position term (ablation) | SPIL · RWF-2000 | accuracy: 88.0 -> 88.7 | Improved |
| use 8 heads instead of 1 head (ablation) | SPIL-Mask · RWF-2000 | accuracy: 88.7 -> 89.2 | Improved |
| use 8 heads instead of 1 head (ablation) | vanilla baseline · RWF-2000 | accuracy: 84.3 -> 84.7 | Improved |
| increase heads from 8 to 64 (ablation) | SPIL-Mask · RWF-2000 | accuracy: 89.2 -> 89.0 | No clear effect |
| set mask threshold d=0.04 instead of 0.01 (ablation) | SPIL-Mask · RWF-2000 | accuracy: 88.7 -> 89.3 | Improved |
| set mask threshold d=0.04 instead of 0.08 (ablation) | SPIL-Mask · RWF-2000 | accuracy: 89.0 -> 89.3 | Improved |
| use SPIL instead of Flow Gated Network on RWF-2000 (ablation) | SPIL-Mask · RWF-2000 | accuracy: 87.3 -> 89.3 | Improved |
| use SPIL instead of Flow Gated Network on Hockey-Fight | SPIL-Mask · Hockey-Fight | accuracy: 98.0 -> 96.8 | Hurt |
| use SPIL instead of Flow Gated Network on Crowd Violence | SPIL-Mask · Crowd Violence | accuracy: 88.8 -> 94.5 | Improved |
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.
- PointConv — PointConv: Deep Convolutional Networks on 3D Point Clouds (CVPR 2019) · 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
- Learning Action Recognition Model From Depth and Skeleton Videos (ICCV 2017 · similarity 0.86)
- SkeleTR: Towards Skeleton-based Action Recognition in the Wild (ICCV 2023 · similarity 0.85)
- Self-Supervised 3D Action Representation Learning With Skeleton Cloud Colorization (TPAMI 2023 · similarity 0.85)
- Learning Multi-View Interactional Skeleton Graph for Action Recognition (TPAMI 2020 · similarity 0.85)
- InfoGCN: Representation Learning for Human Skeleton-Based Action Recognition (CVPR 2022 · similarity 0.85)
- Skeleton Cloud Colorization for Unsupervised 3D Action Representation Learning (ICCV 2021 · 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.
