Reproduce “Representing the Unseen: Rethinking Visual Imagination for Spatial Reasoning” 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

Zengqi Zhao, Zun Wang, Shoubin Yu, Mohit Bansal, Yue Zhang

The paper tests whether spatial reasoning improves when vision-language models receive representations beyond imagined RGB. On IMAGIREP’s 922-question benchmark, authors compare oracle and estimated evidence, including depth, panorama, and top-down geometry. Results vary by question: adding top-down geometry raises one farthest-object score to 75.5%. GEARS scores 61.4% and 77.7%, but does not lead every question type.

Summary by alphaXiv.

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