Researchers have developed a new dual-stream autoregressive framework to address the challenge of creating photorealistic and temporally coherent animatable human avatars from RGB videos. Existing methods often fail to accurately capture cloth dynamics, leading to artifacts or over-smoothed appearances, particularly with out-of-distribution poses. This new approach explicitly models the temporal causality of cloth physics by incorporating both observable geometric information and an implicit internal state. The framework includes a geometric stream for surface displacement propagation and a state stream for fusing current features with historical states, along with motion-adaptive aggregation and adaptive regularization techniques. Experiments show significant improvements in rendering quality, temporal consistency, and generalization to novel motion patterns. AI
IMPACT This research could lead to more realistic and dynamic virtual avatars, impacting fields like gaming, virtual reality, and film production.
RANK_REASON The item describes a new research paper detailing a novel modeling framework for temporal cloth dynamics in avatar creation. [lever_c_demoted from research: ic=1 ai=1.0]
Read on Hugging Face Daily Papers →
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
