InterFusion: Text-Driven 3D Human-Object Interaction Generation Framework
Temel Kavramlar
InterFusion is a two-stage framework for zero-shot 3D human-object interaction generation, significantly outperforming existing methods.
Özet
The study introduces InterFusion, addressing challenges in generating 3D human-object interactions from text descriptions. The framework involves synthesizing anchor poses and optimizing human and object models using spatial constraints. Experimental results demonstrate superior performance over baseline methods.
Structure:
Introduction to the Complex Task of HOI Generation
Challenges Faced in Traditional Approaches
Shift Towards Text-to-3D Methods
Methodology Overview: InterFusion Framework
Two-Stage Approach: Anchor Pose Generation and HOI Scene Generation
Detailed Explanation of Pose-Guided HOI Generation Process
Evaluation of InterFusion Against Baseline Methods
Ablation Studies Demonstrating Importance of Design Choices
InterFusion
İstatistikler
"Our experimental results affirm that Inter-Fusion significantly outperforms existing state-of-the-art methods in 3D HOI generation."
"A total of 235 result prompts are generated, covering most interactions in daily life."
"Our experiments show that the quality of generation can be improved by a large margin and our approach outperforms state-of-the-art HOI generation methods."
Alıntılar
"Our method achieves more stable and higher-quality 3D results under multiple-concept guidance."
"Results demonstrate superior performance over baseline methods."