Fast organization of large photo collections using CUDA

  • Authors:
  • Tim Johnson;Pierre Fite-Georgel;Rahul Raguram;Jan-Michael Frahm

  • Affiliations:
  • Department of Computer Science, University of North Carolina at Chapel Hill;Department of Computer Science, University of North Carolina at Chapel Hill;Department of Computer Science, University of North Carolina at Chapel Hill;Department of Computer Science, University of North Carolina at Chapel Hill

  • Venue:
  • ECCV'10 Proceedings of the 11th European conference on Trends and Topics in Computer Vision - Volume Part II
  • Year:
  • 2010

Quantified Score

Hi-index 0.00

Visualization

Abstract

In this paper, we introduce a system for the automatic organization of photo collections consisting of millions of images downloaded from the Internet. To our knowledge, this is the first approach that tackles this problem exclusively through the use of general-purpose GPU computing techniques. By leveraging the inherent parallelism of the problem and through the use of efficient GPU-based algorithms, our system is able to effectively summarize datasets containing up to three million images in approximately 16 hours on a single PC, which is orders of magnitude faster compared to current state of the art techniques. In this paper, we present the various algorithmic considerations and design aspects of our system, and describe in detail the various steps of the processing pipeline. Additionally, we demonstrate the effectiveness of the system by showing results for a variety of real-world datasets, ranging from the scale of a single landmark, to that of an entire city.