ANT: Adaptive Neural Temporal-Aware Text-to-Motion Model
Fuente:
arXiv
Saved in:
| Main Authors: | , , , , , , , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2025
|
| Subjects: | |
| Online Access: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866914001450958848 |
|---|---|
| author | Chen, Wenshuo Yu, Kuimou Jia, Haozhe Yuan, Kaishen Huang, Zexu Tian, Bowen Lai, Songning Xiao, Hongru Zhang, Erhang Wang, Lei Yue, Yutao |
| author_facet | Chen, Wenshuo Yu, Kuimou Jia, Haozhe Yuan, Kaishen Huang, Zexu Tian, Bowen Lai, Songning Xiao, Hongru Zhang, Erhang Wang, Lei Yue, Yutao |
| contents | While diffusion models advance text-to-motion generation, their static semantic conditioning ignores temporal-frequency demands: early denoising requires structural semantics for motion foundations while later stages need localized details for text alignment. This mismatch mirrors biological morphogenesis where developmental phases demand distinct genetic programs. Inspired by epigenetic regulation governing morphological specialization, we propose **(ANT)**, an **A**daptive **N**eural **T**emporal-Aware architecture. ANT orchestrates semantic granularity through: **(i) Semantic Temporally Adaptive (STA) Module:** Automatically partitions denoising into low-frequency structural planning and high-frequency refinement via spectral analysis. **(ii) Dynamic Classifier-Free Guidance scheduling (DCFG):** Adaptively adjusts conditional to unconditional ratio enhancing efficiency while maintaining fidelity. Extensive experiments show that ANT can be applied to various baselines, significantly improving model performance, and achieving state-of-the-art semantic alignment on StableMoFusion. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2506_02452 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | ANT: Adaptive Neural Temporal-Aware Text-to-Motion Model Chen, Wenshuo Yu, Kuimou Jia, Haozhe Yuan, Kaishen Huang, Zexu Tian, Bowen Lai, Songning Xiao, Hongru Zhang, Erhang Wang, Lei Yue, Yutao Computer Vision and Pattern Recognition While diffusion models advance text-to-motion generation, their static semantic conditioning ignores temporal-frequency demands: early denoising requires structural semantics for motion foundations while later stages need localized details for text alignment. This mismatch mirrors biological morphogenesis where developmental phases demand distinct genetic programs. Inspired by epigenetic regulation governing morphological specialization, we propose **(ANT)**, an **A**daptive **N**eural **T**emporal-Aware architecture. ANT orchestrates semantic granularity through: **(i) Semantic Temporally Adaptive (STA) Module:** Automatically partitions denoising into low-frequency structural planning and high-frequency refinement via spectral analysis. **(ii) Dynamic Classifier-Free Guidance scheduling (DCFG):** Adaptively adjusts conditional to unconditional ratio enhancing efficiency while maintaining fidelity. Extensive experiments show that ANT can be applied to various baselines, significantly improving model performance, and achieving state-of-the-art semantic alignment on StableMoFusion. |
| title | ANT: Adaptive Neural Temporal-Aware Text-to-Motion Model |
| topic | Computer Vision and Pattern Recognition |
| url | https://arxiv.org/abs/2506.02452 |