Reproduce “GameRealm: Grounding Visual Learning in Interactive Games” 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

Weiguang Zhao, Yifan Wang, Ni Yang, Yating Wang, Xingyu Miao, Yankai Jiang, Mulin Yu, Tao Lu and 4 more

GameRealm pairs over 120 million RGB frames from more than 50 games with metric depth and camera parameters, alongside synchronized audio and actions for multi-view and panoramic sequences. In a matched π³ fine-tuning comparison, it reduced Sintel trajectory ATE from 0.0766 to 0.0500. Gains were strongest in camera-pose estimation; depth and reconstruction gains were mixed against the original pretrained model.

Summary by alphaXiv.

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