Research map / Video / Action recognition
CrossGLG: LLM Guides One-shot Skeleton-based 3D Action Recognition in a Cross-level Manner
Tingbing Yan, Wenzheng Zeng, Yang Xiao, Xingyu Tong, Bo Tan, Zhiwen Fang, Zhiguo Cao, Joey Tianyi Zhou
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 CrossGLG to MotionBERT | MotionBERT · NTU 120 one-shot (20 base classes) | accuracy: 35.5 -> 51.0 | Improved |
| add CrossGLG to InfoGCN | InfoGCN · NTU 120 one-shot (100 base classes) | accuracy: 56.1 -> 62.6 | Improved |
| add CrossGLG to HDGCN | HDGCN · NTU 120 one-shot (20 base classes) | accuracy: 39.0 -> 43.0 | Improved |
| add GAP text-guided contrastive method designed for full supervision | InfoGCN · NTU 120 one-shot (20 base classes) | accuracy: 37.0 -> 35.06 | Hurt |
| add CrossGLG to InfoGCN on NTU 60 with 10 base classes | InfoGCN · NTU 60 one-shot (10 base classes) | accuracy: 51.1 -> 57.9 | Improved |
| InfoGCN+CrossGLG vs MotionBERT on NTU 60 with 10 base classes | InfoGCN+CrossGLG · NTU 60 one-shot (10 base classes) | accuracy: 58.3 -> 57.9 | Hurt |
| add CrossGLG to InfoGCN on Kinetics | InfoGCN · Kinetics one-shot (20 base classes) | accuracy: 13.3 -> 17.4 | Improved |
| add global-to-local textual guidance (JID with calibration loss) (ablation) | InfoGCN · NTU 120 one-shot (20 base classes) | accuracy: 37 -> 43.3 | Improved |
| add local-to-global cross-modal interaction (ablation) | InfoGCN · NTU 120 one-shot (20 base classes) | accuracy: 37 -> 42.5 | Improved |
| place JID after the 5th block rather than the 9th (ablation) | InfoGCN + G2L · NTU 120 one-shot (20 base classes) | accuracy: 38.7 -> 43.3 | Improved |
| add text-text self-attention in interaction block (ablation) | InfoGCN + L2G · NTU 120 one-shot (80 base classes) | accuracy: 49.6 -> 60.1 | Improved |
| swap ChatGPT for Gemini or Qwen to generate text descriptions (ablation) | CrossGLG · NTU 120 one-shot (20 base classes) | accuracy: 45.3 / 45.2 / 45.7 | No clear effect |
Closest papers by meaning
- Part-aware Unified Representation of Language and Skeleton for Zero-shot Action Recognition (CVPR 2024 · similarity 0.89)
- Navigating Open Set Scenarios for Skeleton-Based Action Recognition (AAAI 2024 · similarity 0.88)
- Recovering Complete Actions for Cross-dataset Skeleton Action Recognition (NeurIPS 2024 · similarity 0.88)
- Revealing Key Details to See Differences: A Novel Prototypical Perspective for Skeleton-based Action Recognition (CVPR 2025 · similarity 0.88)
- Neuron: Learning Context-Aware Evolving Representations for Zero-Shot Skeleton Action Recognition (CVPR 2025 · similarity 0.88)
- Universal Skeleton Understanding via Differentiable Rendering and MLLMs (ICML 2026 · 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.
