discussion / Drones  / 29 March 2023

ML-ready datasets for aerial/drone wildlife surveys

Groups

I got inspired to collect a list of annotated, appear-to-be-ML-ready datasets related to drone/aerial wildlife surveys:

https://github.com/agentmorris/agentmorrispublic/blob/main/drone-datasets.md

--

Updated: everything below this line is no longer relevant; via a community effort, we verified that we can match annotations to images in all those datasets (and removed a couple where this is not the case).  So, that URL is stable now, and as permanent as something like this gets.  Ergo, everything in this post after this sentence is out of date.

--

I've downloaded all of them to make sure they're accessible and contain at least some kind of annotations, though I haven't actually verified that I can, e.g., match annotations to images.

Don't take a dependency on that URL, just gathering notes before cleaning it up in some more useful way.

Questions for this group:

  1. Does this list already exist somewhere?
  2. If not , what's missing from the list?
  3. Among these datasets, which have you used, and for what?
  4. Among these datasets, which have you *tried* to use, but hit a roadblock?
  5. Are you sitting on well-curated, annotated data that you would like to release were it no so darn big?

For this exercise, I'm focusing specifically on wildlife (take that, vegetation/land cover surveys).

Thanks!

-Dan




Following up here... a few of us huddled virtually and assembled standardized metadata, standardized sample code, and sample annotated images for all the datasets on that list.  I don't know whether this exercise was useful, but it was fun!  

The URL hasn't changed, but I consider the list open for business now, let us know what we're missing:

https://github.com/agentmorris/agentmorrispublic/blob/main/drone-datasets.md