Reproduce “Disen-Forcing: Disentangling Semantic Anchoring from Motion for Autoregressive Video Diffusion” 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
Zhengyang Yu, Akio Hayakawa, Masato Ishii, Qingtao Yu, Xuesong Li, Takashi Shibuya, Yuki Mitsufuji, Jing Zhang
Disen-Forcing adapts autoregressive video diffusion distillation to preserve motion while limiting long-rollout drift. It assigns initial-frame context to sparse attention heads and recent-frame context to most others, then regularizes an attention-based motion proxy. On 30-second Self-Forcing generation, Dynamic Degree is 73.12 and drift is 4.03, versus 40.44 and 7.23 for Self-Forcing; Frame Sink has lower drift, 3.40.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.