Research map / Video / Action recognition
Generative Action Description Prompts for Skeleton-based Action Recognition
Wangmeng Xiang, Chao Li, Yuxuan Zhou, Biao Wang, Lei Zhang
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 generative action-description prompts (GAP) to CTR-GCN | CTR-GCN · NTU RGB+D 120 X-Sub | accuracy (%): 88.9 -> 89.9 | Improved |
| add GAP to ST-GCN backbone (ablation) | ST-GCN · NTU120 X-Sub joint | accuracy (%): 82.6 -> 83.8 | Improved |
| use label name as text prompt (ablation) | GAP · NTU120 X-Sub joint | accuracy (%): 84.6 -> 84.8 | No clear effect |
| use synonym or paragraph prompts (ablation) | GAP · NTU120 X-Sub joint | accuracy (%): 84.6 -> 85.2 | Improved |
| use body part description prompts (ablation) | GAP · NTU120 X-Sub joint | accuracy (%): 84.6 -> 85.4 | Improved |
| combine synonym and body part descriptions (ablation) | GAP · NTU120 X-Sub joint | accuracy (%): 85.4 -> 85.5 | Improved |
| use more body parts and multi-part contrastive loss (ablation) | GAP · NTU120 X-Sub joint | accuracy (%): 85.2 -> 85.4 | Improved |
| increase partition from four to six parts (ablation) | GAP · NTU120 X-Sub joint | accuracy (%): 85.4 -> 85.4 | No clear effect |
| use GPT-3 generated descriptions instead of manual or HAKE descriptions (ablation) | GAP · NTU120 X-Sub joint | accuracy (%): 85.2 -> 85.5 (manual), 85.3 -> 85.5 (HAKE) | Improved |
| use generated prompts instead of learnable prompt vectors (ablation) | GAP · NTU120 X-Sub joint | accuracy (%): 85.2 -> 85.5 | Improved |
| swap CLIP text encoder for BERT (ablation) | GAP · NTU120 X-Sub joint | accuracy (%): 85.2 -> 85.2 | No clear effect |
| use JSD loss instead of KL divergence contrastive loss (ablation) | GAP · NTU120 X-Sub joint | accuracy (%): 85.5 -> 85.7 | Improved |
Closest papers by meaning
- Part-aware Unified Representation of Language and Skeleton for Zero-shot Action Recognition (CVPR 2024 · similarity 0.88)
- Heterogeneous Skeleton-Based Action Representation Learning (CVPR 2025 · similarity 0.88)
- Stitch, Contrast, and Segment: Learning a Human Action Segmentation Model Using Trimmed Skeleton Videos (AAAI 2025 · similarity 0.88)
- Semantic-guided Cross-Modal Prompt Learning for Skeleton-based Zero-shot Action Recognition (CVPR 2025 · similarity 0.88)
- Revealing Key Details to See Differences: A Novel Prototypical Perspective for Skeleton-based Action Recognition (CVPR 2025 · similarity 0.87)
- Neuron: Learning Context-Aware Evolving Representations for Zero-Shot Skeleton Action Recognition (CVPR 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.
