Research map / Video / Video generation
AttT2M: Text-Driven Human Motion Generation with Multi-Perspective Attention Mechanism
Chongyang Zhong, Lei Hu, Zihao Zhang, Shihong Xia
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 body-part attention spatio-temporal encoder (BPST) to VQ-VAE (ablation) | AttT2M · HumanML3D test | FID: 0.213 -> 0.112 | Improved |
| add local motion-word cross-attention (ablation) | AttT2M · HumanML3D test | FID: 0.183 -> 0.112 | Improved |
| add global motion-sentence conditional self-attention (ablation) | AttT2M · HumanML3D test | FID: 0.156 -> 0.112 | Improved |
| add BPST + global-local attention versus vanilla VQ-VAE and transformer (ablation) | AttT2M · HumanML3D test | R Precision Top-1: 0.472 -> 0.499 | Improved |
| add BPST to VQ-VAE reconstruction (ablation) | VQ-VAE · HumanML3D | reconstruction error: 0.6512 -> 0.6427 | Improved |
| AttT2M versus T2M-GPT on KIT-ML | text-to-motion generation · KIT-ML test | FID: 0.717 -> 0.870 | Hurt |
| AttT2M versus T2M-GPT on HumanML3D | text-to-motion generation · HumanML3D test | FID: 0.116 -> 0.112 | Improved |
Later papers that compared with this method
7 later accepted papers report AttT2M as a baseline in their results tables; 7 of them report a better number on most of the results they share with it.
- ActionPlan: Future-Aware Streaming Motion Synthesis via Frame-Level Action Planning (ECCV 2026 · better on 2 of 2 shared results)
- MotionStreamer: Streaming Motion Generation via Diffusion-based Autoregressive Model in Causal Latent Space (ICCV 2025 · better on 2 of 2 shared results)
- Text-to-Any-Skeleton Motion Generation Without Retargeting (ICCV 2025 · better on 1 of 1 shared result)
- Length-Aware Motion Synthesis via Latent Diffusion (ECCV 2024 · better on 4 of 5 shared results)
- ParCo: Part-Coordinating Text-to-Motion Synthesis (ECCV 2024 · better on 2 of 3 shared results)
- MMM: Generative Masked Motion Model (CVPR 2024 · better on 2 of 2 shared results)
- M2D2M: Multi-Motion Generation from Text with Discrete Diffusion Models (ECCV 2024 · better on 2 of 2 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
- Text Motion Translator: A Bi-Directional Model for Enhanced 3D Human Motion Generation from Open-Vocabulary Descriptions (ECCV 2024 · similarity 0.92)
- TM2T: Stochastic and Tokenized Modeling for the Reciprocal Generation of 3D Human Motions and Texts (ECCV 2022 · similarity 0.91)
- Fg-T2M++: LLMs-Augmented Fine-Grained Text Driven Human Motion Generation (IJCV 2025 · similarity 0.91)
- Generating Diverse and Natural 3D Human Motions From Text (CVPR 2022 · similarity 0.90)
- ReMoGPT: Part-Level Retrieval-Augmented Motion-Language Models (AAAI 2025 · similarity 0.89)
- SnapMoGen: Human Motion Generation from Expressive Texts (NeurIPS 2025 · similarity 0.89)
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.
