Reproduce “Geospatial Vector Geometry as a Native Spatial Modality for 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
Huang Fei, Boyi Chen, Hao Xue, Yang Yue, Yehong Xu, Xiaofang Zhou
GeoVec-VLM gives a vision-language model structured vector geometry alongside image regions and language queries. It retains ordered boundary points and paths for query-conditioned spatial reasoning. In a controlled comparison, full GeoVec-VLM raises direction accuracy from 0.704 to 0.804 versus pooled geometry and improves all five task columns. These results are limited to static, aligned 2D scenes.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.