Most state-of-the-art instance segmentation methods produce binary segmentation masks, however, geographic and cartographic applications typically require precise vector polygons of extracted objects instead of rasterized output. This paper introduces PolyWorld, a neural network that directly extracts building vertices from an image and connects them correctly to create precise polygons. The model predicts the connection strength between each pair of vertices using a graph neural network and estimates the assignments by solving a differentiable optimal transport problem. Moreover, the vertex positions are optimized by minimizing a combined segmentation and polygonal angle difference loss. PolyWorld significantly outperforms the state-of-the-art in building polygonization and achieves not only notable quantitative results, but also produces visually pleasing building polygons. Code and trained weights will be soon available on github.