Reproduce “Reconstruction as Spatial Evaluation: Diagnosing Cross-View Spatial Intelligence in Large Vision-Language Models” 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
Junfei Wu, Qiuyue Wang, Xuhong Huang, Qihang Peng, Mingsheng Li, Jian Guan, Qiang Liu, Shu Wu and 3 more
RECONSPATIAL evaluates whether models can reconstruct jointly consistent metric 3D boxes for queried objects across multiple views. Models receive object regions and return boxes in a camera-anchored frame. Structure measures relative layout, while Metric measures physical placement and scale. GPT-6-Astra scores 81.4% Structure and 58.7% Metric, illustrating the gap across three domains; the benchmark does not establish real-world robot performance.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.