Reproduce “VideoMaMa++: Temporally Consistent Video Matting via Preserve-and-Refine Token” 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
Sangbeom Lim, Seoung Wug Oh, Heeji Yoon, Seungryong Kim, Joon-Young Lee
VideoMaMa++ connects short-clip diffusion matting across long videos by passing each clip’s mattes to the next as references and supplying a shared anchor. Its preserve and refine tokens distinguish supplied mattes from coarse masks. On YouTubeMatte at 1920×1080, it reports MAD 1.15 and dtSSD 1.17, versus 1.41 and 1.31 for VideoMaMa; these benchmark results do not establish temporal smoothness.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.