Sharkbook - images are being cloned

Bug Reports
,
1

In which Wildbook did the issue occur? Sharkbook

Not sure if its in relation to us moving towards MiewID but I am finding that encounters are being cloned per fin that is in the photo. Given that Grey Nurse sharks are not identifiable from their fins, and only from their spots, this seems like a glitch.

2

Hi @Sarah_Aus_SpotaShark

I checked a few of your recent encounters to investigate and didn’t find any clones. Can you link me to an encounter and its cloned annotation so I can investigate further?

3

Yeah I was deleting the repeats - ok, when the next image is uploaded and the bug occurs I will resist cleaning up the duplicates and share with you to review the bug.

1 Like
4

Hi @Anastasia finally the bug has reappeared! I promise not to delete so that you can see the cloning thing that is happening.

I uploaded this encounter: https://www.sharkbook.ai/encounters/encounter.jsp?number=6dca36b7-7b60-4b90-a3ad-361cd650db56

And Sharkbook has cloned it here: https://www.sharkbook.ai/encounters/encounter.jsp?number=426729af-3cb0-4e62-9640-8b73635fc32e

This is very frustrating because when you run a search for unapproved encounters you can be fooled into doing the spot mapping on the same encounter twice. You can see it in this list twice with the same sighting ID (e979d53d-7589-4372-b57d-6d69b36e5ad7):

1 Like
5

Thank you! I’ll share this example with the machine learning team. I’ll let you know when I hear back.

6

Here is another example:

https://www.sharkbook.ai/encounters/encounter.jsp?number=659126ce-6410-4db7-aa92-dcd5bc962a09

1 Like
7

In fact, it appears to be cloning many encounters. To the point it makes it really difficult to manage because you cannot work out what is a real encounter versus a fake encounter. Please check here and look down the list and you will see a number of cloned images adding to my list of unapproved images to review: Sharkbook: Wildbook for Sharks | Login

8

Yes I think this is happening on Flukebook too. If there are multiple flukes/dorals assigned to different animals, or if the filename/image is used for multiple whales - say if there are two flukes in the same image.

9

Hi @ShaneGero, This is a separate issue. The sharks @Sarah_Aus_SpotaShark is IDing use spot-mapping, and therefore don’t need separate fin annotations because the identifiable feature on these sharks are primarily their spots, not their fins. A cloned encounter with just a fin annotation isn’t useable for this case.

10

@Sarah_Aus_SpotaShark

We’re still working on this. Thanks for your patience and for the additional examples.

11

You can tell if you’re working from a cloned encounter if you see this in the encounter’s audit trail:

NOTE: cloneWithoutAnnotations([original encounterID]) -> [current encounterID]

Thanks for your patience. We’ve made an update to the sensitivity of the detection model so it should stop cloning fin annotations on Carcharias taurus.

Let us know if you notice any odd behavior like annotations getting missed or if the cloned fins continue and we can adjust the settings again.

This won’t fix the data in the past, but it should limit new issues in the future.

Thank you for the feedback on this! This is a new model, so this kind of feedback helps us refine it so it works better for all users.