Abstract

Recently released open-source pre-trained foundational image segmentation andobject detection models (SAM2+GroundingDINO) allow for geometrically consistentsegmentation of objects of interest in multi-view 2D images. Users can usetext-based or click-based prompts to segment objects of interest withoutrequiring labeled training datasets. Gaussian Splatting allows for the learningof the 3D representation of a scene's geometry and radiance based on 2D images.Combining Google Earth Studio, SAM2+GroundingDINO, 2D Gaussian Splatting, andour improvements in mask refinement based on morphological operations andcontour simplification, we created a pipeline to extract the 3D mesh of anybuilding based on its name, address, or geographic coordinates.

 

Quick Read (beta)