Research map / Vision-language and multimodal / CLIP and zero-shot
SkelHCC: A Hyperbolic CLIP-Driven Cache Adaptation Framework for Skeleton-based One-Shot Action Recognition
Yanan Liu, Anqi Zhu, Jingmin Zhu, Jun Liu, Hossein Rahmani, Mohammed Bennamoun, Farid Boussaid, Dan Xu, Qiuhong Ke
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 |
|---|---|---|---|
| use hyperbolic CLIP instead of Euclidean CLIP (with standard cache) (ablation) | CLIP + Cache · NTU RGB+D 120 (100 base) | accuracy: 62.9 -> 64.8 | Improved |
| use explicit hierarchical hyperbolic CLIP (EH-HCLIP) instead of vanilla HCLIP (ablation) | HCLIP + Cache · NTU RGB+D 120 (100 base) | accuracy: 64.8 -> 67.6 | Improved |
| use LMV-Cache instead of standard cache (ablation) | CLIP (Euclidean) · NTU RGB+D 120 (100 base) | accuracy: 62.9 -> 66.2 | Improved |
| combine EH-HCLIP with LMV-Cache (ablation) | SkelHCC · NTU RGB+D 120 (100 base) | accuracy: 62.9 -> 71.6 | Improved |
| add body-part and body-joint granularities to LMV-Cache (ablation) | SkelHCC · NTU RGB+D 120 (100 base) | accuracy: 67.6 -> 71.6 | Improved |
| add body-part and body-joint granularities to EH-HCLIP (ablation) | SkelHCC · NTU RGB+D 120 (100 base) | accuracy: 68.2 -> 71.6 | Improved |
| use random mask instead of no mask (ablation) | LMV-Cache · NTU RGB+D 120 (100 base) | accuracy: 68.5 -> 66.3 | Hurt |
| use learnable mask instead of no mask (ablation) | LMV-Cache · NTU RGB+D 120 (100 base) | accuracy: 68.5 -> 68.6 | No clear effect |
| use attention mask instead of no mask (ablation) | LMV-Cache · NTU RGB+D 120 (100 base) | accuracy: 68.5 -> 67.1 | Hurt |
| use LLM body-part and body-joint masks (ablation) | LMV-Cache · NTU RGB+D 120 (100 base) | accuracy: 68.5 -> 71.6 | Improved |
| swap GPT-4 for Qwen-3 or Deepseek-v3 to generate masks (ablation) | SkelHCC · NTU RGB+D 120 (100 base) | accuracy: 71.6 -> 71.5 (Qwen-3) | No clear effect |
| use SkelHCC with HD-GCN backbone instead of CrossGLG | SkelHCC · NTU RGB+D 120 (100 base, HD-GCN) | accuracy: 56.8 -> 67.9 | Improved |
Closest papers by meaning
- Boosting Skeleton-based Zero-Shot Action Recognition with Training-Free Test-Time Adaptation (NeurIPS 2025 · similarity 0.90)
- Neuron: Learning Context-Aware Evolving Representations for Zero-Shot Skeleton 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)
- Heterogeneous Skeleton-Based Action Representation Learning (CVPR 2025 · similarity 0.88)
- Skeleton-based Action Recognition with Non-linear Dependency Modeling and Hilbert-Schmidt Independence Criterion (AAAI 2025 · similarity 0.88)
- Revealing Key Details to See Differences: A Novel Prototypical Perspective for Skeleton-based Action Recognition (CVPR 2025 · similarity 0.88)
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.
