MEVA Transcoding FAQ Frequently Anticipated Questions

Changelog:

• v2.0 20jun2021 Updated with post-upload status
• v1.0 02jun2021 Initial release

What data has been updated and why?

We have replaced all ground camera videos (from the school, admin, hospital, and bus nodes) with transcoded versions. The original ("r0") videos were copied directly off the NVR units used during the original collect back in 2018. Some of these files contain errors which prevent software such as ffmpeg from, for example, seeking to random points in the video. The transcoded versions should have fewer errors in general and support seeking in particular.

Following a two-week comment period, the transcoded videos ("r13") were uploaded the evening of June 20, 2021.

Note that any future ground camera videos we release will be recoded in this same format.

Where are the new videos?

The r0 videos were in three s3 "directories": drop-01, drop-02, and drop-03. All transcoded videos are available in a new direcory:

s3://mevadata-public-01/drops-123-r13
They may be downloaded via e.g.

aws s3 sync s3://mevadata-public-01/drops-123-r13 your-local-directory

What format are the transcoded videos?

The transcoded videos are still H.264 in an AVI container.

How were the videos transcoded?

The general process was that each video was broken out into frames; these frames were then recombined back into the transcoded video. The script we used is available. The DIVA performers and evaluation team conducted several rounds of experiments before settling on the final transcoding parameters, which are:

• crf 25 (a libx264 parameter)
• fastdecode (a libx264 tuning parameter)
• 60 frames (2 seconds) between I-frames
• yuv420p pixel format

How does this affect the annotations?

For annotations generated by Kitware for T&E and our 'meva-training' effort, very little if at all. Our annotation process starts with the same process of breaking the video into frames; thus the annotations and transcoded video are starting from the same data source.

How do I know if I'm working with a transcoded video?

All transcoded videos will have an .r13.avi suffix. ("r13" since this was the 13th set of transcoding parameters we tried.)

Will the original clips still be available?

The original clips will be kept, although we do not have plans at this time to make them available via mevadata.org.

How does this affect the ActEV competition?

NIST is already using identically transcoded videos on the Sequestered Data Leaderboard.

Does this affect the UAV videos?

No, only the ground cameras have been transcoded.

Will every video be recoded?

No, a small number were too corrupted to be transcoded. A list of videos which were not transcoded is available here.

Are additional videos being released at this time?

No.

What if I have further questions?

Please post your questions to the meva-data-users group or directly to us at mevadata@kitware.com.