The document summarizes a presentation on carved visual hulls for image-based modeling. It discusses how the method uses a set of images to identify rims around the object's silhouette. It then performs global optimization using graph cuts with the rims as constraints, followed by local refinement. The results show accurate reconstructions on several data sets, with running times of around 2 hours for the global and local steps. Comparisons demonstrate it outperforms other methods at reconstructing concave structures.