Information about the AMOS Dataset

Background

The Archive of Many Outdoor Scenes (AMOS) dataset contains more than 90 million images taken from 24221 webcams located around the world, the vast majority in the United States. Construction of AMOS began in March, 2006 and continues to this day. This dataset is unique in that it contains images from significantly more scenes than in previous work.

The cameras in the dataset were selected by a group of graduate and undergraduate students using a standard web search engine. Images from each camera are captured several times per hour using a custom web crawler that ignores duplicate images and records the capture time. The images from all cameras are 24-bit JPEG files that vary in size. Most are 640 x 480 pixels, but some are at a much higher resolution.

Each image's capture time is recorded in the image's file name. We use the format yyyymmdd_hhmmss.jpg (4 digits for year, and 2 digits each for month, day, hour minute, and second), and the timestamp is in GMT.

Locations of the cameras in AMOS that are located in the continental United States.

We created a large number (currently 170) of montage images that summarize a year of images of a scene. More are available via the AMOS flickr collection.

Why collect so many images?

One impetus for collecting these images was to provide a dataset to enable empirical assessment of our ideas regarding the statistics of natural scenes captured from static cameras. These statistics have not been well studied and are of great importance to surveillance algorithm and application development.

Dataset Access

The dataset is available in two ways:

  1. Through the camera browser, which lists details about each individual camera, as well as an interface to browse the many years of imagery we have archived.
  2. A .zip file archive, where you can download many months of imagery at a time for experimentation. If you plan to download a substantial amount of the AMOS dataset, follow these steps:
    • Download the following script: download_amos.py.
    • Open the file in your favorite text editor.
    • Edit the constants CAMERAS_TO_DOWNLOAD, YEARS_TO_DOWNLOAD, and MONTHS_TO_DOWNLOAD for the segments of the dataset you care about.
    • Run using "python download_amos.py"
    • The files should now start downloading and extracting in your current directory.

There is also a dataset that contains the above, with some ground truth geolocation labels, for use in image geolocation algorithms.

Please cite the following if you use AMOS in your work:

Nathan Jacobs, Nathaniel Roman, Robert Pless, "Consistent Temporal Variations in Many Outdoor Scenes", In IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2007.

Publications

  • Nathan Jacobs, Nathaniel Roman, Robert Pless, "Consistent Temporal Variations in Many Outdoor Scenes", In IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2007. [pdf]
  • Nathan Jacobs, Scott Satkin, Nathaniel Roman, Richard Speyer, Robert Pless, "Geolocating Static Cameras", In IEEE International Conference on Computer Vision (ICCV), 2007. [pdf]
  • Nathan Jacobs, Nathaniel Roman, Robert Pless, "Toward Fully Automatic Geo-Location and Geo-Orientation of Static Outdoor Cameras", In IEEE Workshop on Applications of Computer Vision (WACV), 2008. [pdf]
  • Nathan Jacobs, Robert Pless, "Calibrating and Using the Global Network of Outdoor Webcams", In ACM/IEEE International Conference on Distributed Smart Cameras (ICDSC), 2009.
  • Nathan Jacobs, Richard Souvenir, Robert Pless, "The Global Webcam Imaging Network", In Applied Imagery Pattern Recognition Workshop (AIPR), 2009.
  • Nathan Jacobs, Walker Burgin, Richard Speyer, David Ross, Robert Pless, "Adventures in Archiving and Using Three Years of Webcam Images", In IEEE CVPR Workshop on Internet Vision, 2009.[pdf]
  • Nathan Jacobs, Walker Burgin, Nick Fridrich, Austin Abrams, Kylia Miskell, Bobby H. Braswell, Andrew D. Richardson, Robert Pless, "The Global Network of Outdoor Webcams: Properties and Applications", In ACM International Conference on Advances in Geographic Information Systems (SIGSPATIAL GIS), 2009. [pdf]
  • Austin Abrams, Nick Fridrich, Nathan Jacobs, Robert Pless, "Participatory Integration of Live Webcams into GIS", In International Conference on Computing for Geospatial Research and Applications (COM.GEO), 2010. [pdf]
  • Austin Abrams, Emily Feder, Robert Pless. "Exploratory Analysis of Time-Lapse Imagery with Fast Subset PCA", in IEEE Workshop on Applications of Computer Vision (WACV) 2011. [pdf]
  • Nathan Jacobs, Kylia Miskell, Robert Pless. "Webcam Geo-localization using Aggregate Light Levels", in IEEE Workshop on Applications of Computer Vision (WACV) 2011.
  • Austin Abrams, Robert Pless. "Web-Accessible Geographic Integration and Calibration of Webcams", in ACM Transactions on Multimedia Computing, Communication, and Applications 9(1); 8, 2013.
  • Austin Abrams, Christopher Hawley, Robert Pless. "Heliometric Stereo: Shape from Sun Position", in European Conference on Computer Vision (ECCV), 2012. [pdf]
  • Austin Abrams, Kylia Miskell, Robert Pless. "The Episolar Constraint: Monocular Shape from Shadow Correspondence", in IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2013. [pdf]
  • Calvin Murdock, Nathan Jacobs, Robert Pless. "Webcam2Satellite: Estimating Cloud Maps from Webcam Imagery", in IEEE Workshop on Applications of Computer Vision (WACV) 2013. [pdf]

People

The following people have contributed to the collection, presentation, and analysis of the AMOS dataset.

Acknowledgements

This project is supported under NSF IIS 0546383: "CAREER: Passive Vision, What Can Be Learned by a Stationary Observer". Any opinions, findings, and conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the views of the National Science Foundation. This project also gratefully acknowledges AWS Convergence Technologies Inc. for allowing us to archive a collection of images from the weatherbug camera network.