Research map / Video / Action recognition
Neural Koopman Pooling: Control-Inspired Temporal Dynamics Encoding for Skeleton-Based Action Recognition
Xinghan Wang, Xin Xu, Yadong Mu
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 |
|---|---|---|---|
| Replace temporal average pooling with Koopman pooling | CTR-GCN · NTU RGB+D 120 X-Sub | accuracy: 88.9 -> 90.0 | Improved |
| Replace temporal average pooling with Koopman pooling | CTR-GCN · NTU RGB+D 60 X-View | accuracy: 96.8 -> 96.8 | No clear effect |
| Replace temporal average pooling with Koopman pooling | CTR-GCN · NW-UCLA | accuracy: 96.5 -> 97.0 | Improved |
| Match sequences using DMD-derived exemplar dynamics rather than embedding-space distance | CTR-GCN one-shot recognition · NTU RGB+D 120 | one-shot accuracy: 65.6 -> 68.1 | Improved |
| Match sequences using DMD-derived exemplar dynamics rather than embedding-space distance | CTR-GCN one-shot recognition · NW-UCLA | one-shot accuracy: 83.3 -> 89.9 | Improved |
| Add eigenvalue normalization to Koopman pooling (ablation) | CTR-GCN · NTU RGB+D 120 X-Sub, joint | accuracy: 85.6 -> 85.7 | Improved |
| Add eigenvalue normalization to Koopman pooling (ablation) | CTR-GCN · NTU RGB+D 120 X-Sub, bone | accuracy: 87.1 -> 87.2 | Improved |
| Add eigenvalue normalization to Koopman pooling (ablation) | CTR-GCN · NTU RGB+D 120 X-Sub, joint motion | accuracy: 82.5 -> 82.7 | Improved |
| Add eigenvalue normalization to Koopman pooling (ablation) | CTR-GCN · NTU RGB+D 120 X-Sub, bone motion | accuracy: 83.0 -> 83.1 | Improved |
| Add Koopman pooling (ablation) | CTR-GCN · NTU RGB+D 120 X-Sub, joint and bone modalities | accuracy: 84.9 -> 85.7; 85.7 -> 87.2 | Improved |
| Add Koopman pooling (ablation) | CTR-GCN · NTU RGB+D 120 X-Sub, joint motion and bone motion | accuracy: 81.4 -> 82.7; 81.2 -> 83.1 | Improved |
| Remove stride in temporal convolution (ablation) | CTR-GCN · NTU RGB+D 120 X-Sub | accuracy: 84.9 -> 84.6 | Hurt |
Closest papers by meaning
- Action Recognition with Dynamic Image Networks (TPAMI 2017 · similarity 0.88)
- Discriminatively Learned Hierarchical Rank Pooling Networks (IJCV 2017 · similarity 0.87)
- InfoGCN++: Learning Representation by Predicting the Future for Online Skeleton-Based Action Recognition (TPAMI 2024 · similarity 0.87)
- Neuron: Learning Context-Aware Evolving Representations for Zero-Shot Skeleton Action Recognition (CVPR 2025 · similarity 0.86)
- Revealing Key Details to See Differences: A Novel Prototypical Perspective for Skeleton-based Action Recognition (CVPR 2025 · similarity 0.85)
- Heterogeneous Skeleton-Based Action Representation Learning (CVPR 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.
