Hey there :)

You could use Timelapse (for data annotation and automatic metadata extraction, e.g. date/time, temperature, etc) in combination with Megadetector (it works with videos too). You basically run the Megadetector through your batch of videos and then use the output json file in Timelapse (to filter out empty videos or automate the annotation process). 

Here's the documentation for Timelapse:

https://saul.cpsc.ucalgary.ca/timelapse/pmwiki.php?n=Main.HomePage

And here you can find detailed info on how to integrate the Megadetector input into Timelapse.

https://saul.cpsc.ucalgary.ca/timelapse/uploads/Guides/TimelapseImageRecognitionGuide.pdf

I hope this is useful.

Cheers,

Lucy

You are correct that at its core, MegaDetector only knows about still images.  However, we have found that it's quite reasonable to run MegaDetector on something like every 10th frame of a video (that's still typically three frames per second) to determine whether animals are present.  I would say video represents around 10% of total MegaDetector use right now.  If you have very small animals that flash in and out of frame quickly, this probably isn't a good approach (this comes up a lot with video surveys for birds), but for most animals, this has worked great.

As Lucy highlights, most MegaDetector users load their MegaDetector results into Timelapse to review their images; Timelapse lets you say, for example, "show me only the images that MegaDetector thinks are not empty".  It can do the same for videos.  But it's not necessary to use Timelapse; some users, for example, review their videos in dedicated behavioral coding software and just want to get blanks out of the way first.  MegaDetector can help with that.

All that said, it's a little bit more involved to process videos than to process stills, so we recommend you reach out to us at [email protected], and we can run a few test videos, make sure things are working OK, then walk you through the process.  For the adventerous, the relevant script is here, and the notebook we use to do this for batch jobs is here, but for new users, we typically recommend starting by email.

And just for fun, here is a video of MegaDetector running on a video of a raccoon.  Because raccoons.

Zamba Cloud can handle video!

Hey there - we can help you, please send us some of the videos and will build a pipeline for it 

aakash at thinkevolvelabs dot com 

- Aakash Gupta