Bad identified re-sightings

Support
,
1

Hello!

Since 2018, I have been recording Salamandra salamandra in Amphibian and Reptile Wildbook (https://amphibian-reptile.wildbook.org/).

I have now recorded approximately 950 photos in various projects.

An important goal is to determine whether the found salamanders have been sighted before. This is intended, for example, to identify possible survivors of the Bsal disease.

Initially, the quality of the identified re-sightings was satisfactory. Now – seemingly with the increasing number of animals recorded in the project – the results are very poor.

For example, in my Bsal - Ratiger Wald project, I have now recorded 248 sightings.

Of the 27 new sightings from November 1, 2025, the program could only detect one re-sighting. At times, I displayed up to 50 suggestions – without success.

Visually, however, I was able to identify 5 definite re-sightings. I added them manually, but this leaves the program far from being a practical application.

Are there any suggestions on how to improve this?

Are other users experiencing similar problems?

Is the problem—as I suspect—due to the increasing number of entries, or do you think there’s another reason?

The photos are of good quality!

Regards, Paul_b

2

Hi @Paul_Bachhausen

It would help if you could link to match page examples where the correct candidate isn’t shown, but you know it to be an individual that already exists in ARW so we can research this.

3

For example:

Does this help?

Manually connected:

https://amphibian-reptile.wildbook.org/individuals.jsp?id=e6e0ebdc-f3bd-4a4d-8987-51ebb6f7ede9

Amphibian and Reptile Wildbook

1 Like
4

Thanks for the example! It looks like there’s no location ID set on either encounter.

Screenshot 2025-12-03 at 10.57.56 AM
Screenshot 2025-12-03 at 10.57.56 AM440×260 14.6 KB

Matching works by looking for annotations of the same species in the same location ID. When the location ID is missing, Wildbook checks the entire database for matches, where false positives can push the correct match further down the results (especially when there is only a single encounter of the correct individual to look for).

Did your previously uploaded encounters from when you had stronger match results have location IDs on them? If so, it could explain why your current match results aren’t as accurate.

Let me know if you need me to add a new location to ARW for you by submitting a Feature Request post and filling in the new location template so I have what I need to get this added.

5

Thanks for the reply.

I didn’t have a location ID before either.

But since I’m only searching within a single project, a location ID shouldn’t be necessary, right?

6

That’s helpful context. It does look like the top candidates in the latest match all share the same leafy background. I wonder if the matching algorithm is focusing more on the background than the markings on the salamander. There’s also a possibility that the annotation on the original encounter of ID-01246 needs be redrawn and re-matched since it’s cropped parts of the head and tail:

Screenshot 2025-12-03 at 10.51.48 AM
Screenshot 2025-12-03 at 10.51.48 AM548×492 108 KB

I’m linking the original match result for the November encounter here so I can compare if correcting the annotation in the previous encounter makes it move up in the match results when it’s run again: Amphibian and Reptile Wildbook

Can you start a new match here (Image menu > Start another match) and then when the match is ready, use the project filter on the match page to select your project? I’m hoping that the corrected annotation on the older encounter makes it appear when you restart the match on your newer encounter.

7

I tried it. Unfortunately, no improvement.

8

Thanks for checking! I’ll need to discuss with my teammates. In the meantime, feel free to share additional examples as you come across them.

9

Thank you so much for trying to help.

Here are a few that Wildbook hadn’t found:

https://amphibian-reptile.wildbook.org/individuals.jsp?id=f85bac4c-61c7-4304-b64e-73c44b053707

https://amphibian-reptile.wildbook.org/individuals.jsp?id=1daa078e-dd1f-4cd4-8be3-1bd5eff8eadd

https://amphibian-reptile.wildbook.org/individuals.jsp?id=e6e0ebdc-f3bd-4a4d-8987-51ebb6f7ede9

10

and:

https://amphibian-reptile.wildbook.org/individuals.jsp?id=f323c2dc-bda7-42f5-ad30-4b4ee781ddd0

https://amphibian-reptile.wildbook.org/individuals.jsp?id=7d523f44-0470-4beb-905b-5e78564e10a8

11

Thanks for the additional examples! We’re actively researching this and I hope to have an update for you soon.

1 Like
12

Hi,

I might chime in!

Here are some examples of encounters that I have identified as recaptures but were not identified by the matching algorithm (see also: Amphibian & Reptile Wildbook Low Recapture Recognition Rate - #3 by Anastasia )

2 Likes
13

Thank you! The machine learning team is looking into this. We suspect that the rotation isn’t being considered by the matching algorithms. We are working to get you unblocked as soon as we can.

1 Like
14

Hi @Paul_Bachhausen and @Max.Muehlenhaupt

Thanks for your patience. You should be seeing improved MiewID results for Salamandra salamandra now. Synchronization for annotation rotation has been fixed and we’ve seen improvement to match results in testing. Feel free to resend any questionable match results back through ID. Let us know if you see this arise again.

2 Likes
15

Great, thank you!

1 Like
16

Hi Anastasia,

yes, I tested it and it’s much better now.

Thank you so much!

Paul

1 Like
17

Thank you both for your detailed reports!

18

I have one more request.
During the revision process, the admin created two sightings that I cannot delete.
Please delete these:

https://amphibian-reptile.wildbook.org/occurrence.jsp?number=e0c9cc55-4a21-4369-b8f6-910fbcf98329

https://amphibian-reptile.wildbook.org/occurrence.jsp?number=fd8c734a-87a8-443f-8376-618aa5c791c4

19

I’ve deleted the linked encounters and sightings.

20

Thanks