核心概念
Proposing GPT-Connect for training-free scene-aware text-driven human motion generation.
摘要
The content introduces GPT-Connect, a framework for generating scene-aware human motion sequences without the need for extensive training data. It addresses the limitations of existing methods that focus on blank backgrounds rather than 3D scenes. The framework leverages ChatGPT to connect a motion generator with 3D scenes in a training-free manner. Key highlights include:
- Introduction to text-driven human motion generation.
- Challenges of generating motions in 3D scenes.
- Proposal of GPT-Connect framework.
- Utilization of ChatGPT as a connector between motion generator and 3D scenes.
- Detailed explanation of Scene-GPT and GPT-Generator channels.
- Evaluation on HUMANISE dataset with quantitative and qualitative results.
统计
この論文は、HUMANISEデータセットでの定量的評価を行っています。
引用
"Extensive experiments demonstrate the efficacy and generalizability of our proposed framework."