Reproduce “Matrix-Game 3.0: Real-Time and Streaming Interactive World Model with Long-Horizon Memory” 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
Zile Wang, Zexiang Liu, Jiaxing Li, Kaichen Huang, Baixin Xu, Fei Kang, Mengyin An, Peiyu Wang and 15 more
Matrix-Game 3.0 generates interactive video views and uses earlier views to preserve scene details on return. It retrieves historical frames based on camera pose and field-of-view overlap, combining them with recent frames. In qualitative action-reversal examples, returning views recover structures and details; the system reports up to 40 frames per second at 720p with a 5B model.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.