Researchers have developed GNOCHI, a new generative model designed to create realistic 3D human-human interactions in virtual environments. This model utilizes a conditional variational autoencoder (cVAE) to generate poses for one person based on the pose of another, addressing the challenge of data scarcity with an automated supervised data augmentation strategy. GNOCHI also incorporates a self-supervised loss function that uses volumetric proxies to resolve collisions and ensure physically accurate interactions, which current state-of-the-art methods struggle to achieve. AI
IMPACT Enables more realistic and controllable human-human interactions in virtual environments, potentially impacting VR/AR and simulation.
RANK_REASON Academic paper detailing a new generative model for human-human interactions. [lever_c_demoted from research: ic=1 ai=1.0]
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
