Reproduce “SGF+: Decoupling Gradient Flows for Autoregressive Video Generation” 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
Zihan Su, Junhao Zhuang, Yaowei Li, Siwen Lu, Haoran Li, Lingen Li, Haoyu Wu, Weiyang Jin and 5 more
Self Gradient Forcing Plus (SGF+) separates the parameters used to write key–value memory from those used to denoise in autoregressive video generation, while keeping both roles connected through causal attention and a shared future-generation objective. In a 240-second chunkwise test, it improves four VBench scores over SGF, though Dynamic Degree is slightly lower; 24-hour examples are not a quantitative benchmark.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.