Reproduce “L3G-World: Real-Time Infinite World Generation with Latent 3D Gaussian 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
Jinhyeok Choi, JoungBin Lee, Jongmin Lee, Minkyeong Jeon, Jiyoung Kim, Kaeun Han, Yoonjin Yoon, Seungryong Kim
L3G-World generates interactive video worlds using a revisable 3D Gaussian scene memory that stores video-VAE features. It renders this memory into latent conditioning for each video chunk, then updates it by adding missing content and refining visible Gaussians. On DL3DV loop-closure tests, it achieves 16.00 self-consistency PSNR versus 13.96 for Spatia; Matrix-Game 3.0 has lower LPIPS, and tests use static scenes.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.