Datasets:

---

facebook

/

SA-FARI

Like 38

Follow

AI at Meta 15.2k

Thanks for releasing the SA-FARI dataset. It is an exciting resource, especially since there are very few large scale camera trap video datasets available.

We're currently using SA-FARI in our research and would appreciate some clarification on a few aspects of the dataset.

1. I'm trying to take the subset of videos described in the paper as coming from arboreal/canopy camera traps in Costa Rica (~2,790 videos, mounted 8–24m up). Is there a way to identify or filter those samples using the available metadata?

2. How were empty or false trigger videos handled during dataset curation? Is there any way to obtain access to empty videos, for research purposes?

3. Approximately what percentage of the original raw video collection was empty before filtering?

I'd appreciate any clarification.
Thanks again for making this dataset publicly available!