SuperLectures.com

SEARCHING IN ONE BILLION VECTORS: RE-RANK WITH SOURCE CODING

Full Paper at IEEE Xplore

Image and Video Indexing and Retrieval

Presented by: Hervé Jégou, Author(s): Hervé Jégou, INRIA, France; Romain Tavenard, University of Rennes 1, France; Matthijs Douze, INRIA, France; Laurent Amsaleg, CNRS, France

Recent indexing techniques inspired by source coding have been shown successful to index billions of high-dimensional vectors in memory. In this paper, we propose an approach that re-ranks the neighbor hypotheses obtained by these compressed-domain indexing methods. In contrast to the usual post-verification scheme, which performs exact distance calculation on the short-list of hypotheses, the estimated distances are refined based on short quantization codes, to avoid reading the full vectors from disk. We have released a new public dataset of one billion 128-dimensional vectors and proposed an experimental setup to evaluate high dimensional indexing algorithms on a realistic scale. Experiments show that our method accurately and efficiently re-ranks the neighbor hypotheses using little memory compared to the full vectors representation.


  Speech Transcript

|

  Slides

Enlarge the slide | Show all slides in a pop-up window

0:00:16

  1. slide

0:00:28

  2. slide

0:01:32

  3. slide

0:02:53

  4. slide

0:06:22

  5. slide

0:08:02

  6. slide

0:08:42

     5. slide

0:08:48

     6. slide

0:09:47

  7. slide

0:10:52

  8. slide

0:12:05

  9. slide

0:12:59

 10. slide

0:14:02

 11. slide

  Comments

Please sign in to post your comment!

  Lecture Information

Recorded: 2011-05-25 16:15 - 16:35, Club A
Added: 14. 6. 2011 22:17
Number of views: 61
Video resolution: 1024x576 px, 512x288 px
Video length: 0:17:52
Audio track: MP3 [6.03 MB], 0:17:52