Reproduce “SpatialCLI: Learning to Reason With Spatial Tools, Then Without Them” 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
Yang Zhou, Zixuan Huang, Sunzhu Li, Zhuo Yang, Chen Zhang, Shunian Chen, Caijun Yan, Jianyao Xu and 5 more
SpatialCLI equips a vision-language model with localization, segmentation, depth, and pose tools, then trains it on tool interactions to improve both tool use and direct answers. On SpatialCLI-Bench, Qwen3-VL-8B-Instruct scores 35.3% without tools and 66.5% with them; SpatialCLI-8B scores 72.7% and 91.3% after training. These benchmark results concern perception-centric tasks.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.