Research map / Vision-language and multimodal / Human and embodied modeling
Beyond Global Alignment: Fine-Grained Motion-Language Retrieval via Pyramidal Shapley-Taylor Learning
Hanmo Chen, Guangtao Lyu, Chenghao Xu, Jiexi Yan, Xu Yang, Cheng Deng
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 STI distillation loss (lambda_S=1 vs 0) (ablation) | PST · HumanML3D text-to-motion | R@10: 39.12 -> 48.22 | Improved |
| increase lambda_S from 1 to 2 (ablation) | PST · HumanML3D text-to-motion | R@10: 48.22 -> 45.54 | Hurt |
| add joint-to-segment self-distillation loss (lambda_D=1 vs 0) (ablation) | PST · HumanML3D text-to-motion | R@10: 32.41 -> 48.22 | Improved |
| increase lambda_D from 1 to 3 (ablation) | PST · HumanML3D text-to-motion | R@10: 48.22 -> 35.49 | Hurt |
| lower compression ratio to 0.1 (from 0.25) (ablation) | PST · HumanML3D text-to-motion | R@10: 48.22 -> 31.67 | Hurt |
| raise compression ratio to 0.75 (from 0.25) (ablation) | PST · HumanML3D text-to-motion | R@10: 48.22 -> 31.76 | Hurt |
| use MotionPatch representation instead of Guo feature (ablation) | PST · HumanML3D text-to-motion | R@10: 30.14 -> 48.22 | Improved |
| add STI distillation loss (lambda_S=1 vs 0) (ablation) | PST · KIT-ML text-to-motion | R@10: 51.18 -> 57.87 | Improved |
| add self-distillation loss (lambda_D=1 vs 0) (ablation) | PST · KIT-ML text-to-motion | R@10: 42.10 -> 57.87 | Improved |
| increase lambda_S from 1 to 2 (ablation) | PST · KIT-ML text-to-motion | R@10: 57.87 -> 55.79 | No clear effect |
| replace Guo feature with MotionPatch representation (ablation) | PST · KIT-ML text-to-motion | R@10: 39.89 -> 57.87 | Improved |
| PST fine-grained pyramidal alignment versus prior best (Lyu et al.) | HumanML3D All protocol · text-to-motion | R@1: 11.80 -> 12.45 | Improved |
Closest papers by meaning
- SGAR: Structural Generative Augmentation for 3D Human Motion Retrieval (NeurIPS 2025 · similarity 0.91)
- Chronologically Accurate Retrieval for Temporal Grounding of Motion-Language Models (ECCV 2024 · similarity 0.90)
- MG-MotionLLM: A Unified Framework for Motion Comprehension and Generation across Multiple Granularities (CVPR 2025 · similarity 0.89)
- ReMoGPT: Part-Level Retrieval-Augmented Motion-Language Models (AAAI 2025 · similarity 0.89)
- Tri-Modal Motion Retrieval by Learning a Joint Embedding Space (CVPR 2024 · similarity 0.88)
- HMVLM:Human Motion-Vision-Language Model via MoE LoRA (NeurIPS 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.
