Research map / Video / Action recognition
Semantics-Guided Neural Networks for Efficient Skeleton-Based Human Action Recognition
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 joint-type semantics to graph learning (ablation) | SGN joint-level module · NTU60 CS and CV | accuracy: 86.9 -> 87.5; 92.8 -> 93.7 | Helped |
| Add joint-type semantics to message passing (ablation) | SGN joint-level module · NTU60 CS and CV | accuracy: 86.9 -> 88.6; 92.8 -> 94.1 | Helped |
| Also use joint-type semantics in graph learning when it is already used for message passing (ablation) | SGN joint-level module · NTU60 CS and CV | accuracy: 88.6 -> 88.7; 94.1 -> 94.1 | No clear effect |
| Add frame-index semantics (ablation) | SGN without temporal convolution · NTU60 CS and CV | accuracy: 86.8 -> 87.8; 92.8 -> 93.7 | Helped |
| Add frame-index semantics alongside temporal convolution (ablation) | SGN frame-level module · NTU60 CS and CV | accuracy: 88.7 -> 89.0; 94.1 -> 94.5 | Helped |
| Replace per-frame hierarchical graph modeling with one global graph across joints and frames (ablation) | SGN · NTU60 CS and CV | accuracy: 87.3 -> 89.0; 93.3 -> 94.5 | Hurt |
| Add spatial MaxPooling to aggregate joints within each frame (ablation) | SGN · NTU60 CS and CV | accuracy: 88.3 -> 89.0; 93.9 -> 94.5 | Helped |
| Add data augmentation (ablation) | Strong baseline · NTU60 CS and CV | accuracy: 79.2 -> 80.6; 81.4 -> 87.1 | Helped |
| Add velocity features (ablation) | Strong baseline with data augmentation · NTU60 CS and CV | accuracy: 80.6 -> 85.3; 87.1 -> 91.4 | Helped |
| Replace AveragePooling with MaxPooling (ablation) | Strong baseline with data augmentation and velocity · NTU60 CS and CV | accuracy: 85.3 -> 86.9; 91.4 -> 92.8 | Helped |
| Add semantics (joint type and frame index) (ablation) | SGN w/o Sem. · NTU60 CS and CV | accuracy: 86.9 -> 89.0; 92.8 -> 94.5 | Helped |
| Add semantics (joint type and frame index) (ablation) | SGN w/o Sem. · NTU120 C-Subject and C-Setup | accuracy: 77.4 -> 79.2; 79.2 -> 81.5 | Helped |
Closest papers by meaning
- Skeleton-Based Action Recognition With Directed Graph Neural Networks (CVPR 2019 · similarity 0.91)
- Heterogeneous Skeleton-Based Action Representation Learning (CVPR 2025 · similarity 0.91)
- InfoGCN: Representation Learning for Human Skeleton-Based Action Recognition (CVPR 2022 · similarity 0.91)
- Dynamic Semantic-Based Spatial Graph Convolution Network for Skeleton-Based Human Action Recognition (AAAI 2024 · similarity 0.90)
- Skeleton-based Action Recognition with Non-linear Dependency Modeling and Hilbert-Schmidt Independence Criterion (AAAI 2025 · similarity 0.89)
- Revealing Key Details to See Differences: A Novel Prototypical Perspective for Skeleton-based Action Recognition (CVPR 2025 · similarity 0.89)
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.