Dex2HOI: Dexterous Bimanual Two-Object Interaction Generation
Dex2HOI is a diffusion model that generates realistic, dexterous movements for two hands interacting with up to two objects simultaneously from text. It achieves this by using a Dual-Stream Diffusion approach coordinated via cross-attention and fusing the outputs through a Motion Fusion Network, enabling state-of-the-art performance for complex bimanual manipulation.