Research map / Video / Video generation
TMR: Text-to-Motion Retrieval Using Contrastive 3D Human Motion Synthesis
Mathis Petrovich, Michael J. Black, Gül Varol
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 jointly trained motion synthesis (reconstruction) branch to contrastive training (ablation) | TMR · KIT-ML (b) All with threshold | R@3: 36.87 -> 41.93 | Improved |
| use InfoNCE instead of margin-based contrastive loss (ablation) | TMR · KIT-ML (b) All with threshold | R@3: 34.46 -> 41.93 | Improved |
| add contrastive loss to TEMOS (ablation) | TEMOS · KIT-ML (b) All with threshold | R@3: 30.84 -> 41.93 | Improved |
| filter wrong negatives with text similarity above 0.8 (ablation) | TMR · KIT-ML (b) All with threshold | R@3: 36.02 -> 41.93 | Improved |
| set temperature tau to 1.0 instead of 0.1 (ablation) | TMR · KIT-ML (b) All with threshold | R@3: 41.93 -> 3.61 | Hurt |
| increase batch size from 32 to 128 (ablation) | TMR · KIT-ML (b) All with threshold | R@3: 41.93 -> 36.75 | Hurt |
| vary contrastive loss weight lambda_NCE away from 0.1 (ablation) | TMR · KIT-ML (b) All with threshold | R@3: 41.93 -> 34.46 at 1.0 | Hurt |
| TMR versus Guo et al. on HumanML3D (all with threshold) | text-to-motion retrieval · HumanML3D (b) | MedR: 54.00 -> 19.00 | Improved |
| TMR versus Guo et al. on easy small-batch protocol at R@10 | text-to-motion retrieval · HumanML3D (d) Small batches | R@10: 96.58 vs 95.36 | Mixed |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- TEMOS — TEMOS: Generating Diverse Human Motions from Textual Descriptions (ECCV 2022) · better on 7 of 7 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Tri-Modal Motion Retrieval by Learning a Joint Embedding Space (CVPR 2024 · similarity 0.90)
- VimoRAG: Video-based Retrieval-augmented 3D Motion Generation for Motion Language Models (NeurIPS 2025 · similarity 0.89)
- ReMoGPT: Part-Level Retrieval-Augmented Motion-Language Models (AAAI 2025 · similarity 0.89)
- Towards Detailed Text-to-Motion Synthesis via Basic-to-Advanced Hierarchical Diffusion Model (AAAI 2024 · similarity 0.88)
- AMD: Autoregressive Motion Diffusion (AAAI 2024 · similarity 0.87)
- Light-T2M: A Lightweight and Fast Model for Text-to-motion Generation (AAAI 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.
