Research map / Video / Video generation
Motion-Aligned Word Embeddings for Text-to-Motion Generation
Ke Han, Yueming Lyu, Nicu Sebe
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 Lself (ablation) | MATE with MoMask · HumanML3D | FID: 0.339 -> 0.040 | Improved |
| add Lcross (ablation) | MATE with MoMask · HumanML3D | FID: 0.044 -> 0.040 | Improved |
| add Lalign (ablation) | MATE with MoMask · HumanML3D | FID: 0.049 -> 0.040 | Improved |
| add Lsent (ablation) | MATE with MoMask · HumanML3D | Top-1: 0.324 -> 0.550 | Improved |
| add Lrec (ablation) | MATE with MoMask · HumanML3D | FID: 0.042 -> 0.040 | Improved |
| fine-tune only word embedding layers instead of all CLIP layers (ablation) | MATE with MoMask · HumanML3D | Top-1: 0.014 -> 0.550 | Improved |
| fine-tune only word embedding layers instead of subsequent layers (ablation) | MATE with MoMask · HumanML3D | FID: 7.611 -> 0.040 | Improved |
| use word embedding fine-tuning instead of LoRA adapter (ablation) | MATE with MoMask · HumanML3D | FID: 0.051 -> 0.040 | Improved |
| train word embeddings instead of no trainable layers (ablation) | MATE with MoMask · HumanML3D | FID: 0.045 -> 0.040 | Improved |
| add Gaussian attention prior over whole sequence (ablation) | MATE with MoMask · KIT | FID: 0.253 -> 0.197 | Improved |
| use Gaussian prior over whole sequence instead of binary segment mask (ablation) | MATE with MoMask · KIT | FID: 0.217 -> 0.197 | Improved |
| replace CLIP with MATE-enhanced CLIP in MoMask | MATE with MoMask · KIT | FID: 0.204 -> 0.197 | No clear effect |
Closest papers by meaning
- MoLingo: Motion-Language Alignment for Text-to-Human Motion Generation (CVPR 2026 · similarity 0.90)
- Chronologically Accurate Retrieval for Temporal Grounding of Motion-Language Models (ECCV 2024 · similarity 0.89)
- TM2T: Stochastic and Tokenized Modeling for the Reciprocal Generation of 3D Human Motions and Texts (ECCV 2022 · similarity 0.89)
- LaMP: Language-Motion Pretraining for Motion Generation, Retrieval, and Captioning (ICLR 2025 · similarity 0.88)
- AToM: Aligning Text-to-Motion Model at Event-Level with GPT-4Vision Reward (CVPR 2025 · similarity 0.87)
- MG-MotionLLM: A Unified Framework for Motion Comprehension and Generation across Multiple Granularities (CVPR 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.
