Duplicate Encounters

[Kaci]

What Wildbook are you working in?
Snail Wildbook

What is the entire URL out of the browser, exactly where the error occurred?

Wildbook

Wildbook

Wildbook

Can you describe what the issue is you’re experiencing?
I am not sure if restarting the 1000 detections caused this problem, but the encounters seemed to have been duplicated and separated by images. For instance, this encounter b269f616-05e2-41c5-b2a5-49387c766731 was duplicated 3 times for each individual image:

The 3 duplicate encounters:
7bc92133-048d-4640-a346-5164315b7f7a
06607e49-f585-4c12-91fe-547e6b6efe96
d72a8079-3e78-4fa4-9807-ab2607d7a3ab

Also some of the detection and identification only works for the 3 duplicated encounters. However, the original encounter with all 3 images will sometimes say " there was an error with task b926c4f2-df2d-42f7-ba5d-0e479b56aa55: No data to match against, please refine your matching parameters." when trying to run the match detection.

There are other encounters with this same problem too. If there is any way to put the respective images into its corresponding encounter and delete the duplicated ones, please let me know.

Can you provide steps on how to reproduce what you’re experiencing?
On My Encounters page, I clicked on encounters that were created on 2025-02-06 and compared it to encounters created on 2023-03-30 (and other dates too) and found there were some encounters duplicated and separated by image.

[Anastasia]

Hi @Kaci

Thanks for letting us know. Detection models predict multiple overlapping bounding boxes and we use a technique called non-maximum suppression (NMS) to narrow it down to one box. Sometimes our chosen model NMS value allows in extra bounding boxes, and this is an example of that.

In your encounter examples, we found two annotations in an image which forced the generation of a second Encounter. Because an encounter represents one animal (and two bounding boxes mean two animals detected), we clone the encounter to hold the second bounding box. This is really important behavior for social species but not here.

I’ll talk to the ML team about increasing the NMS value and correcting the existing detections. I’ll follow up with you if I need any other info to help resolve this.

[Anastasia]

Thanks for your patience!

We’ve changed the NMS from 0.4 to 0.5. You can delete the cloned encounters to clean up the duplicates. You can tell when an encounter is a clone of the original if you see this in the Audit trail:

Let us know if the adjustment helps or if you’re starting to see missed annotations so we can refine it again.