StreamOPD: A Post-Training Recipe with Spatio-Temporal Cue Gating for Streaming Video Understanding
UniX Lab
non-profit
AI & ML interests
Unifying vision, reasoning the world, empowering agents. UniX Lab stands at the frontier intersection of multimodal intelligence. With unified modeling as our cornerstone, we expand from visual generation to deep reasoning and autonomous agents. We believe in the power of collaboration and invite you to join this journey of exploring the boundaries of intelligence.
Recent Activity
Papers
StreamOPD: A Post-Training Recipe with Spatio-Temporal Cue Gating for Streaming Video Understanding
WorldReasonBench: Human-Aligned Stress Testing of Video Generators as Future World-State Predictors