Reproduce “Learning the Autoregressive: Direct Score-Difference Estimation with Amortized Future Credit for 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
Zixuan Duan, Yabo Chen, Xunzhi Xiang, Xin Zhang, Haibin Huang, Yin Chen, Chi Zhang, Qi Fan
Leta refines an initialized few-step video generator by directly estimating the teacher–student correction and adding an estimated signal from a later chunk. In controlled runs with five shared seeds, Leta scores 84.5 on VBench Overall versus DMD’s 83.6, and 2.348 on VideoAlign Overall versus 1.692. Text alignment is lower than DMD’s, and FutureCredit is a biased approximation.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.