Reproduce “Beyond Appearance: Evidence-Guided Programmatic 3D Scene Reconstruction with Agentic Lighting 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

Zeting Luo, Yifan Li, Shuai Yang

LightHarness reconstructs an editable, executable 3D scene from one photograph. It combines geometric, intrinsic, and illumination cues with persistent memory of light-source hypotheses to guide revisions to geometry, materials, or lighting. On a 32-image evaluation, it reports 15.03 dB PSNR and 0.216 DreamSim versus VIGA’s 14.28 dB and 0.242, but trails on SSIM in this test set.

Summary by alphaXiv.

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