ANT: Adaptive Neural Temporal-Aware Text-to-Motion Model

Abstract

This paper proposes ANT (Adaptive Neural Temporal-Aware Text-to-Motion Model), which significantly improves semantic alignment and generation efficiency in text-to-motion tasks. By introducing semantic temporal-aware modules and Dynamic Classifier-Free Guidance (DCFG) strategy, our model better understands and generates temporally and semantically coherent motions.

Key Contributions

• Temporal Awareness: Novel semantic temporal-aware module design
• Dynamic Guidance: DCFG strategy for improved generation control
• Enhanced Alignment: Better semantic alignment between text and motion
• Improved Efficiency: Faster generation with maintained quality

Technical Highlights

• Temporal-aware neural architecture
• Dynamic classifier-free guidance mechanism
• Improved semantic understanding for motion generation
• Efficient training and inference pipeline

Download paper here

Recommended citation: Jia, H. et al. (2024). "ANT: Adaptive Neural Temporal-Aware Text-to-Motion Model." arXiv preprint arXiv:2506.02452. https://arxiv.org/abs/2506.02452