✦
Summary
DuoMatching is a few-step video-generation framework built around joint and marginal distribution matching. Joint supervision targets consistent autoregressive rollouts, while marginal supervision improves frame quality and semantic alignment.
Key Features
Few-step video generation.
Joint distribution matching.
Marginal frame supervision.
Streaming rollouts.
Semantic alignment.
Visual-quality preservation.
Released model resources.
Public inference code.
About DuoMatching
Video-model researchers can evaluate distilled streaming generation with released code and model resources. Core capabilities include Few-step video generation, Joint distribution matching, Marginal frame supervision, and Streaming rollouts.
Public code or model weights are linked from the project, with setup instructions for running or adapting the system. Additional characteristics include Semantic alignment, Visual-quality preservation, Released model resources, and Public inference code.