4 papers · 1 filter
Panoptic Scene Program Diffusion Transformer
Chika Maduabuchi
Modern text-to-image models produce high-fidelity images but still struggle with compositional prompts that require instance identity, attribute ownership, counting, spatial orderi…
Event-Driven Video Generation
Chika Maduabuchi, Jindong Wang
Current text-to-video models can make individual frames look convincing while still getting simple interactions wrong: objects move before contact, an intended action is skipped, a…
Temporal Pair Consistency for Variance-Reduced Flow Matching
Chika Maduabuchi, Jindong Wang
Continuous-time generative models, such as diffusion models, flow matching, and rectified flow, learn time-dependent vector fields but are typically trained with objectives that tr…
Entropy-Controlled Flow Matching
Chika Maduabuchi
Modern vision generators transport a base distribution to data through time-indexed measures, implemented as deterministic flows (ODEs) or stochastic diffusions (SDEs). Despite str…