Reproduce “Proxy2World: Learning to Generate Worlds From Lightweight Proxies without Seeing Them” 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
Hongli Xu, Weilong Yan, Anbang Wang, Chunyu Zou, Siyu Hong, Jingwei Huang
Proxy2World uses simplified scene geometry to guide video generation, grounding early denoising in proxy depth before jointly refining RGB video and depth while retaining camera guidance. Trained on ordinary posed RGBD video without authored proxy–video pairs, it reaches 0.6795 geometric alignment and 64.18% adherence win rate on 78 ProxyBench sequences, but Cosmos-Depth has higher adherence.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.