Pose Diversity

Pose diversity research focuses on generating and controlling the variety of human poses in digital environments, addressing challenges in accurately representing and manipulating 3D human movement from limited data. Current research employs diffusion models and flow-based generative models, often incorporating incremental evolution strategies and multi-hypothesis aggregation to improve pose realism and handle uncertainty inherent in 2D-to-3D projection. This work is significant for advancing virtual and augmented reality applications, particularly in areas like digital human interaction, avatar generation, and realistic human insertion into scenes, improving the fidelity and naturalness of human representation in these contexts.