Reproduce “GROME: GROunding Factorized MEmory into Video World Models” 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

Fang Yu, Xiaoxue Wu, Haiyu Zhang, Yangfu Li, Botong Zhao, Yu Qiao, Tianfan Xue, Yue Lu and 1 more

GROME adds separate memories for persistent scenery and moving entities to frozen LingBotV2. It stores local 3D pointmaps and retrieves references for each target view without fine-tuning. On dynamic-scene revisits, it reports static-region PSNR of 24.111 versus 21.243 and entity consistency of 0.769 versus 0.691 against temporal retrieval; motion accuracy is not clearly improved in a 200-scene evaluation.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.