Large camera trap project help

What Wildbook are you working in?

I’m starting a project with ~50,000 photo sequences of chital deer. Each sequence has multiple photos (average 41 photos per sequence).

I’ve been chatting with Anastasia about ways to manage this, and she suggested reaching out to the broader community for help. Hopefully someone with a lot of experience uploading camera trap photos can help!

My first big problem is that we have far too many photos to deal with. If each encounter = 1 photo, I’m looking at >2 million encounters. It seems that my unit of work should be on the sightings basis, but it’s not possible to edit sightings easily. I think one solution is to upload photo sequences/sightings as encounters. Assuming a sequence of photos = 1 deer, this might be reasonable. At this stage, I’m thinking of only uploading sequences with 1 deer, and waiting to deal with multi-deer sequences later.
Does anyone have experience setting up their uploads to have one trigger/sequence of photos = one encounter?
Even then, there are a bunch of issues that I’m not sure how to handle.
If a sequence of photos = 3 deer photos and 2 blank photos, is there a good/easy way to delete any photo that doesn’t have a boundary box?
Is there an easy way to quickly and efficiently delete blurry, poor quality, photos that have no hope of producing a match? I have far more data than I need, so being able to select the good photos would be really helpful.
Matches seem to only happen on a photo-to-photo basis. If I match one photo of deer #001 with one photos of deer #123, I’d like all photos of #001 and #123 to have the same name/label.
How have other people handled that issue?
Anyway, if anyone has some tips/suggestions, I’d be appreciative for any help. At this stage, the task of dealing with this large of a dataset looks almost impossible in the current system, so I’m hoping someone knows some tricks to help out.
Thank you,

Welcome @ben!

Tagging some Community contributors who work with camera traps that might be able to weigh in:

Hi Ben,

We’re facing similar challenges at Lynx for Wildbook and are unsure how to handle them effectively.

We’ve been considering manually removing photos from the sequences before uploading, keeping only the most informative image from each sequence. However, when more than two lynxes appear in a sequence, things get complicated.

We’d love to hear if anyone else has found a solution that works well.


My goal is to reduce the amount of manual work. removing photos prior to upload would be way too many man-hours.
When you have two lynx, are they generally mother-offspring?
I’m hoping to upload only single deer sequences at first, but we’re also really interested in demographics, so the timing/survival of fawns is really important. If I could find a good way to deal with two deer sequences, that would be really helpful as well.

Hi ben,

Sounds like an interesting project! We set our cameras to just take one photo instead of the 3-photo burst, so I don’t have any advice there. However, we have had really nice success using machine learning to filter out blank or low quality images. Our codes are trained for Kenyan species though but that could be an option if there is some open source code avaliable?