Reproduce “SpatialCast: Reading the 3D World via Cast Shadows” 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
Enhan Li, Jinghao Zhou, Jiaze Zhang, Hongyang Du
SpatialCast tests whether multimodal models can use shadows to reason about how physical scene changes alter images. It evaluates scene-image matching, image prediction after interventions, and change inference from paired images. In a controlled light-change test, GPT-6.1 Sol reaches 81.7% shadow-only agreement versus 25.0% without shadows, a diagnostic score that may include indistinguishable choices on these questions.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.