Reproduce “MindView: Benchmarking Generative Multi-View Spatial Intelligence” 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
Yanbo Wang, Siqi Luo, Siyu Zhang, Chenming Zhu, Wenzhao Zheng, Qian Yu, Juncheng Yan, Bin Fu and 4 more
MindView evaluates whether models can generate a requested viewpoint in static indoor scenes from language instructions and partial images. Its 1,500-sample benchmark covers camera motions and viewpoints inferred from objects, relations, and reference cues. More context did not improve results: target coverage rose from 88.4% to 92.0% from six to ten views, while Overall fell. Fine-tuning BAGEL raised Overall from 40.0 to 64.4.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.