FastMap: Revisiting Dense and Scalable Structure from Motion

2025-05-07

AI Summary

To address the scalability bottlenecks of global Structure-from-Motion (SfM) methods—such as COLMAP and GLOMAP—in large-scale scenarios, including poor parallelizability and superlinear optimization complexity (growing with the number of image matches), this paper introduces the first fully GPU-accelerated global SfM framework. Our method achieves efficient large-scale reconstruction through three key innovations: (1) GPU-accelerated dense matching for preprocessing; (2) an incremental pose graph optimization algorithm with linear time complexity; and (3) a lightweight, robust reprojection error minimization scheme. Crucially, the overall optimization complexity scales linearly with the number of image pairs, independent of the numbers of keypoints or 3D points. Extensive experiments on large-scale scenes demonstrate that our framework achieves 10–100× speedup over COLMAP and GLOMAP while maintaining comparable camera pose accuracy.

Technology Category

• Computer Vision: Motion & Tracking
• Search and Optimization: Distributed Search
• Machine Learning: Learning with Manifolds

Application Category

• Graph Algorithms and Modeling for the Web: Representation, reconstruction, and subgraph or motif discovery in Web-related graphs
• Search and Retrieval-Augmented AI: Efficiency and scalability of Web search engines
• Security and Privacy: Large-scale security measurements

Abstract

We propose FastMap, a new global structure from motion method focused on speed and simplicity. Previous methods like COLMAP and GLOMAP are able to estimate high-precision camera poses, but suffer from poor scalability when the number of matched keypoint pairs becomes large. We identify two key factors leading to this problem: poor parallelization and computationally expensive optimization steps. To overcome these issues, we design an SfM framework that relies entirely on GPU-friendly operations, making it easily parallelizable. Moreover, each optimization step runs in time linear to the number of image pairs, independent of keypoint pairs or 3D points. Through extensive experiments, we show that FastMap is one to two orders of magnitude faster than COLMAP and GLOMAP on large-scale scenes with comparable pose accuracy.

Problem

• Improves scalability of Structure from Motion methods
• Addresses poor parallelization in existing SfM techniques
• Reduces computational cost of optimization steps

Innovation

• GPU-friendly operations for parallelization
• Linear-time optimization steps
• Faster than COLMAP and GLOMAP