Multiview Extended Video with Activities (MEVA) Dataset README

Overview

The MEVA dataset was collected as part of the Intelligence Advanced Research Projects Activity (IARPA) DIVA program. It is designed to support performers on the DIVA program and the broader research community focused on activity detection in data from simultaneous, multi-camera environments.

The data was collected using a variety of commercial, off-the-shelf cameras to replicate equipment used in typical real-world environments. The data may therefore include irregularities not seen in high-end, research-grade video data. Camera G639 has been rotated. We are planning a transition to transcoded data; see the transcoding FAQ for more details.

Access

The MEVA Known Facility Dataset 1 ("KF1") is approximately 330 hours of video across 4001 video clips totalling 470 GB. It is available via Amazon Simple Storage Service (S3) via a no-cost download sponsored by Amazon's AWS Public Dataset Program. The s3 bucket name is mevadata-public-01.

Several command line tools and GUI clients are available for downloading from S3, e.g. s3cmd, available at https://s3tools.org/s3cmd. Once installed, commands similar to

$ aws s3 sync s3://mevadata-public-01/drops-123-r13 .
...will synchronize the ground camera 'drops-123-r13' directory into the current directory.

Kitware does not endorse or warrant the utility of any particular S3 client. Your use of Amazon S3 is subject to Amazon's Terms of Use. The accessibility of the MEVA KF1 data from Amazon S3 is provided "as is" without warranty of any kind, expressed or implied, including, but not limited to, the implied warranties of merchantability and fitness for a particular use. Please do not contact Kitware for assistance with Amazon services.

Directory Structure & Filenames

The original ground camera data was staged in three drops. As of 20 June 2021, these videos have been transcoded and are available in a single directory; see the transcoding FAQ for more details. The data is available in:
  • drops-123-r13 : 3956 clips / 444 GB / 324.2 hours
  • uav-drop-01 : 45 clips / 26 GB / 4.6 hours
The directory organization follows a video/facility/date/hour/video clip structure. Ground camera video files are typically five minutes in length, and filenames have the following structure:

YYYY-MM-DD.timestamp-start.timestamp-end.camera-location.camera-number
For example, the filename 2018-03-07.16-50-00.16-55-00.admin.G329.r13.avi is parsed as:
  • Recorded started on March 7, 2018 starting at 16:50:00.
  • Recording ended at 16:55:00.
  • The camera was located in/on the admin building (see metadata, below).
  • The camera number was G329.
  • This clip has been transcoded per the 'r13' process.
The video data is organized into a four-level hierarchy of facility id, date, then hour, then videos. Ground-camera data was recorded on several NAS units whose clocks were synchronized via GPS. The NAS software was configured to record five-minute clips; however, clips do not all necessarily start or stop on even five-minute boundaries. A few clips may be shorter than five minutes due to transmission errors or collection event anomalies. Some clips were collected at the beginning of the event and may be shorter than five minutes.

UAV data was collected by a pair of DJI drones at 3840x2160 @ 30fps; more details can be found in the UAV data readme at https://s3.amazonaws.com/mevadata-public-01/uav-drop-01/meva-uav-drop-01-readme.pdf.

Dataset updates

  • 20 June 2021: The ground camera videos have been transcoded; see the transcoding FAQ for more details.
  • 29 April 2020: Two dataset updates:
  • A block of 130 video clips from camera G328 were identified as corrupted, and have been removed. The clips are from 2018-03-05, between 10:10:16 and 11:23:54am.
  • Camera G639 was mounted sideways. The 83 videos on AWS have been replaced with rotated versions so that up in the world is also up in the video.
The complete list of affected videos may be found at http://mevadata.org/resources/update-20200429.txt.

MEVA Known Facility Definitions

This 1.0 release of the MEVA dataset releases the MEVA Known Facility set 1 (KF1). Please see https://mevadata.org for more details.

License & Citation

The "Multiview Extended Video with Activities" (MEVA) dataset by Kitware Inc. and the Intelligence Advanced Research Projects Activity (IARPA) is licensed under a Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0.) See LICENSE-MEVA-dataset.txt for the full license, available at https://mevadata.org/resources/MEVA-data-license.txt.

The dataset is described in our WACV 2021 paper. The bibtex citation is:

@InProceedings{Corona_2021_WACV,
    author    = {Corona, Kellie and Osterdahl, Katie and Collins, Roderic and Hoogs, Anthony},
    title     = {MEVA: A Large-Scale Multiview, Multimodal Video Dataset for Activity Detection},
    booktitle = {Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision (WACV)},
    month     = {January},
    year      = {2021},
    pages     = {1060-1068}
}

Acknowledgment

The Multiview Extended Videos with Activities (MEVA) dataset collection work is supported by Intelligence Advanced Research Projects Activity contract number 2017-16110300001.

Changelog

  • 20-jun-2021: Updated with revised directories after transcoding.
  • 02-jun-2021: HTML version; updated with citation information and transcoding faq.
  • 29-apr-2019: Updated for section 3.1
  • 12-dec-2019: Updated for drop-03
  • 06-nov-2019: Updated for UAV drop-01
  • 30-sep-2019: Updated with sample download commands
  • 24-sep-2019: Updated for drop-02
  • 06-sep-2019: Updated for AWS Public Dataset Program access
  • 21-may-2019: Adapted for mevadata.org
  • 25-mar-2019: Initial release