This project explores the capabilities and limitations of Large Reconstruction Models (LRMs)
for building reconstruction from street-view imagery. It addresses the unique challenges
of field-of-view coverage, occlusions and entanglement in street-view imagery. This is explored
through training-free multi-model inference combining large-scale pre-trained foundation models,
including instance segmentation, diffusion-based inpainting for handling occlusions, and
single-view reconstruction techniques.