Towards Internet-scale multi-view stereo

TLDR

The paper proposes a method to extend existing multi‑view stereo techniques to very large, unstructured photo collections. It decomposes the collection into overlapping photo groups processed in parallel, solves the clustering as a constrained optimization, and merges the resulting reconstructions with a parallel, out‑of‑core algorithm that filters low‑quality results and enforces global visibility constraints. Applied to several large Flickr datasets, including one with over ten thousand images, the approach produced a 3D reconstruction of nearly thirty million points.

Abstract

This paper introduces an approach for enabling existing multi-view stereo methods to operate on extremely large unstructured photo collections. The main idea is to decompose the collection into a set of overlapping sets of photos that can be processed in parallel, and to merge the resulting reconstructions. This overlapping clustering problem is formulated as a constrained optimization and solved iteratively. The merging algorithm, designed to be parallel and out-of-core, incorporates robust filtering steps to eliminate low-quality reconstructions and enforce global visibility constraints. The approach has been tested on several large datasets downloaded from Flickr.com, including one with over ten thousand images, yielding a 3D reconstruction with nearly thirty million points.