Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
Zero-Shot Open-Vocabulary Human Motion Grounding with Test-Time Training
Yunjiao Zhou, Xinyan Chen, Junlang Qian, Lihua Xie, Jianfei Yang
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 test-time training with LLM sub-action decomposition (LSP) over no TTT with rule-based segmentation (ablation) | ZOMG · HumanML3D grounding | mAP: 5.86 -> 22.69 | Improved |
| add LLM-guided Language Semantic Partition instead of rule-based segmentation (ablation) | ZOMG · HumanML3D grounding | mAP: 5.86 -> 10.16 | Improved |
| add learnable soft mask [M] in SMO (ablation) | ZOMG · HumanML3D grounding | mAP: 22.69 -> 38.63 | Improved |
| add smoothness loss to soft mask (ablation) | ZOMG · HumanML3D grounding | mAP: 38.63 -> 41.54 | Improved |
| add exclusivity loss to soft mask (ablation) | ZOMG · HumanML3D grounding | mAP: 38.63 -> 47.28 | Improved |
| add both smoothness and exclusivity losses (ablation) | ZOMG · HumanML3D grounding | mAP: 47.28 -> 50.46 | Improved |
| use ZOMG instead of AdaZAD | motion grounding · HumanML3D grounding | mAP: 41.77 -> 50.46 | Improved |
| use ZOMG instead of AdaZAD | motion grounding · BABEL grounding | mAP: 10.15 -> 13.44 | Improved |
| optimize only soft masks with frozen encoder instead of adapter parameters | test-time training efficiency · HumanML3D grounding (TTT, 100 steps) | Samples/s: 6.65 -> 23.25 | Improved |
| augment retrieval training data with ZOMG-grounded segments | MESM · HumanML3D retrieval protocol A (T2M) | T2M retrieval: 19.20 -> 19.38 | Improved |
| augment retrieval training data with ZOMG-grounded segments | MESM · KIT-ML retrieval protocol B (T2M) | T2M retrieval: 43.04 -> 42.04 | Hurt |
| add more TTT optimization steps (ablation) | ZOMG · HumanML3D grounding | grounding error | Improved |
Closest papers by meaning
- Go to Zero: Towards Zero-shot Motion Generation with Million-scale Data (ICCV 2025 · similarity 0.89)
- VideoGEM: Training-free Action Grounding in Videos (CVPR 2025 · similarity 0.89)
- OMG: Towards Open-vocabulary Motion Generation via Mixture of Controllers (CVPR 2024 · similarity 0.87)
- Part-aware Unified Representation of Language and Skeleton for Zero-shot Action Recognition (CVPR 2024 · similarity 0.87)
- HMVLM:Human Motion-Vision-Language Model via MoE LoRA (NeurIPS 2025 · similarity 0.87)
- Does Video-Text Pretraining Help Open-Vocabulary Online Action Detection? (NeurIPS 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.
