Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
SA-DVAE: Improving Zero-Shot Skeleton-Based Action Recognition by Disentangled Variational Autoencoders
Sheng-Wei Li, Zi-Xiang Wei, Wei-Jie Chen, Yi-Hsin Yu, Chih-Yuan Yang, Jane Yung-jen Hsu
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 feature disentanglement (semantic-related and irrelevant heads) (ablation) | SA-DVAE · NTU-60 ZSL random split (55/5) | accuracy (%): 69.26 -> 82.21 | Improved |
| add feature disentanglement (ablation) | SA-DVAE · NTU-120 ZSL random split (110/10) | accuracy (%): 39.73 -> 49.18 | Improved |
| add adversarial total correlation penalty on top of disentanglement (ablation) | SA-DVAE · NTU-60 ZSL random split (55/5) | accuracy (%): 82.21 -> 84.20 | Improved |
| add adversarial total correlation penalty (ablation) | SA-DVAE · PKU-MMD ZSL random split (46/5) | accuracy (%): 60.97 -> 66.54 | Improved |
| add total correlation penalty on top of FD (ablation) | SA-DVAE · NTU-60 GZSL random split (55/5) | harmonic mean H (%): 70.71 -> 75.27 | Improved |
| add feature disentanglement (ablation) | SA-DVAE · PKU-MMD ZSL random split (46/5) | accuracy (%): 60.13 -> 60.97 | No clear effect |
| feed disentangled z^r_x instead of f_x to seen classifier (ablation) | SA-DVAE · NTU-60 GZSL random split (55/5) | harmonic mean H (%): 75.27 -> 71.74 | Hurt |
| add pose canonicalization to skeleton data (ablation) | SA-DVAE · NTU-60 GZSL random split (55/5) | harmonic mean H (%): 75.27 -> 71.31 | Hurt |
| add pose canonicalization to skeleton data (ablation) | SA-DVAE · NTU-60 ZSL random split (55/5) | accuracy (%): 84.20 -> 84.03 | No clear effect |
| add LLM-augmented class descriptions | SA-DVAE · NTU-120 ZSL random split (110/10) | accuracy (%): 50.67 -> 57.16 | Improved |
| use ST-GCN + CLIP instead of ST-GCN + Sentence-BERT features (ablation) | SA-DVAE · NTU-60 ZSL random split (55/5) | accuracy (%): 74.38 -> 84.20 | Improved |
| increase KL weights beta_x and beta_y to 0.2 (ablation) | SA-DVAE · NTU-60 random split | GZSL harmonic mean H (%): 75.27 -> 67.05 | Hurt |
Later papers that compared with this method
1 later accepted paper reports SA-DVAE as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- Boosting Skeleton-based Zero-Shot Action Recognition with Training-Free Test-Time Adaptation (NeurIPS 2025 · better on 13 of 13 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
- Semantic-guided Cross-Modal Prompt Learning for Skeleton-based Zero-shot Action Recognition (CVPR 2025 · similarity 0.89)
- Part-aware Unified Representation of Language and Skeleton for Zero-shot Action Recognition (CVPR 2024 · similarity 0.88)
- Neuron: Learning Context-Aware Evolving Representations for Zero-Shot Skeleton Action Recognition (CVPR 2025 · similarity 0.88)
- SCD-Net: Spatiotemporal Clues Disentanglement Network for Self-Supervised Skeleton-Based Action Recognition (AAAI 2024 · similarity 0.87)
- Boosting Skeleton-based Zero-Shot Action Recognition with Training-Free Test-Time Adaptation (NeurIPS 2025 · similarity 0.87)
- Skeleton-based Action Recognition with Non-linear Dependency Modeling and Hilbert-Schmidt Independence Criterion (AAAI 2025 · similarity 0.87)
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.
