Reproduce “RepForcing: Representation-Space Initialization for Autoregressive Video Distillation” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Siming Fu, Zheming Fu, Lianrui Mu, Haoyang Huang, Chaojie Gu, Ruizhe He
RepForcing initializes causal video generation by training a student to match clean video features from a frozen video model, rather than distilling an intermediate teacher’s trajectories. After DMD, in four-step chunk-wise evaluation, it reports VBench Total of 84.66 versus 84.04 and VisionReward of 10.074 versus 6.326, with Semantic lower than Causal Forcing.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.