Research map / Video / Action recognition
Vision-Language Action Knowledge Learning for Semantic-Aware Action Quality Assessment
Huangbiao Xu, Xiao Ke, Yuezhou Li, Rui Xu, Huanqi Wu, Xiaofeng Lin, Wenzhong Guo
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 |
|---|---|---|---|
| plug MVLA (textual semantics via CLIP) into TSA | TSA · FineDiving (w/ DN) | AIoU@0.75: 34.31 -> 67.16 | Improved |
| plug MVLA into six existing AQA methods | USDL, MUSDL, CoRe, TSA, TPT, HGCN · MTL-AQA (w/o DD) | Spearman rho: 0.8861 -> 0.9142 (USDL); 0.9431 -> 0.9529 (TPT) | Improved |
| plug MVLA into five methods on JIGSAWS | USDL, MUSDL, CoRe, TPT, HGCN · JIGSAWS | Avg Spearman rho: 0.84 -> 0.87 (CoRe) | Improved |
| plug MVLA into five methods on Fis-V (quality-grade text) | MS-LSTM, GDLT, CoRe, TPT, MLP-Mixer · Fis-V | Avg MSE: 14.01 -> 13.06 (MLP-Mixer) | Improved |
| add frame-level, stage-level and video-level similarity together (ablation) | MVLA VLKL · FineDiving (w/ DN) | Spearman rho: 0.9340 -> 0.9419 | Improved |
| use only video-level similarity (ablation) | MVLA VLKL · FineDiving (w/ DN) | Spearman rho: 0.9340 -> 0.9243 | Hurt |
| add VLKL (text alignment) on top of CLIP-Image features (ablation) | TSA · FineDiving (w/ DN) | AIoU@0.5: 88.22 -> 92.52 | Improved |
| add SCA on top of CLIP-Image features (ablation) | TSA · FineDiving (w/ DN) | R-l2 (x100): 0.3248 -> 0.3178 | Improved |
| replace I3D-only features with added CLIP-Image visual features (ablation) | TSA · FineDiving (w/ DN) | Spearman rho: 0.9203 -> 0.9266 | Improved |
| add quality descriptions to multi-level texts (ablation) | MVLA · FineDiving (w/ DN) | Spearman rho: 0.9419 -> 0.9438 | No clear effect |
| freeze CLIP text encoder (ablation) | MVLA · FineDiving (w/ DN) | Spearman rho: 0.9419 -> 0.9390 | Hurt |
| use stronger backbone (VST or ViFi-CLIP) (ablation) | MVLA · FineDiving (w/ DN) | Spearman rho: 0.9419 -> 0.9457 (VST) | Improved |
Closest papers by meaning
- Human-Centric Fine-Grained Action Quality Assessment (TPAMI 2025 · similarity 0.90)
- MoAKE: Toward Unified All-in-One Action Quality Assessment via Mixture of Action Knowledge Experts (ECCV 2026 · similarity 0.90)
- FineParser: A Fine-grained Spatio-temporal Action Parser for Human-centric Action Quality Assessment (CVPR 2024 · similarity 0.89)
- Kronecker Mask and Interpretive Prompts are Language-Action Video Learners (ICLR 2025 · similarity 0.88)
- FACT: Frame-Action Cross-Attention Temporal Modeling for Efficient Action Segmentation (CVPR 2024 · similarity 0.87)
- Unsupervised Action Segmentation via Fast Learning of Semantically Consistent Actoms (AAAI 2024 · 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.
