Researchers have developed StreamHOI, a novel framework designed for real-time, long-duration human-object interaction (HOI) video generation. Unlike previous methods limited to offline, short-video generation, StreamHOI addresses the challenges of maintaining interaction awareness within a low-latency streaming system. The framework employs bias-guided memory-specialized training and a memory distance scaling module to optimize how the generator accesses and utilizes historical memory, particularly for HOI regions and surrounding contexts. Evaluations show StreamHOI significantly improves interaction plausibility, object fidelity, and human quality while achieving a speed of 17.6 FPS with a 0.75-second first-chunk latency. AI
IMPACT Enables real-time interactive applications by improving the efficiency and plausibility of long-duration HOI video generation.
RANK_REASON The cluster contains an academic paper detailing a new framework for video generation. [lever_c_demoted from research: ic=1 ai=1.0]
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
