After uploading new data via Bulk Import to check for recaptures of fire salamanders, the match results are predominantly suggesting individuals from the most recent dataset . These are mainly animals that I recorded on the same day.
In contrast, there is very little matching against my older data or previously recorded individuals . As a result, individuals that I have already recorded in the past and that I can recognize from their individual markings in the images are often not presented as possible matches based on the annotation.
Hi @Steve.Naumann
Is this the import you’re referring to? Which encounter day is overly represented in the match candidates and which match result links are they? I spot checked a few encounters and I’m seeing a variety of individuals suggested across different days (such as here: Wildbook).
Hi Anastasia,
Thank you for your quick response!
Yes, this is the import I am currently working on. You are right that for some encounters, different individuals are suggested and everything works as expected. However, for others, only individuals from the same bulk import are being suggested. This has not occurred nearly as frequently with the other bulk imports I have worked with so far.
The screenshots illustrate the issue I am referring to. The individuals being suggested, such as DD_RB_455, are also from the same import. In screenshots 2 and 3, for example, you can see the same individual (DD_RB_75), but this individual is not suggested as a match in screenshot 2. I hope this helps what I mean. Thank you very much!
Here is the additional screenshot from the previous message.
Thanks! Can you link to those match pages for me so I can troubleshoot?
here´re some links. Thanks for your support.
Amphibian and Reptile Wildbook
Amphibian and Reptile Wildbook
This one has no match at all
(We couldn’t find any matches based on your current filters. Please try adjusting them or consider this a new individual)
Amphibian and Reptile Wildbook
Thanks for the links!
Do you see how the encounter annotation is missing most of the salamander? The way annotations work is that identification only looks at what’s inside that bounding box to find a match. Since a small portion of the back is all that’s selected, the match results aren’t very good. When you see an annotation like this, you’ll need to manually edit the annotation and then start a new match to see appropriate results. After I fixed the annotation, the quality of the match candidates improved:
This one is not as extreme as the previous example, but the direction of the annotation matters, too.
This screenshot is taken from the edit annotation workflow from the encounter this match came from. The handle on this annotation should point to the animal’s direction of travel. Here, it’s aiming left, but it should be pointing up. This helps make sure that the orientation of the patterns it compares to are consistent. After editing it so that it’s pointing where the salamander is facing, the match candidates have more variety: Wildbook
This is also the same issue as the above examples. The annotation on the salamander doesn’t contain most of the salamander in it.
After I fixed the annotation and started a new match, I can see match candidates now: Wildbook
When you review your uploads, it’s important to also review your annotations as well. The detection algorithm does a pretty good job, but it can still make mistakes. I did see a few of your other encounters as match candidates that also had questionable annotations that I fixed and resent to identification. Once you finish auditing your encounters from your recent import and fix any bad annotations, you should start seeing better quality match results.
If the annotation is good and you’re still not seeing a match that you feel should be there, feel free to start a new match and add the Hotspotter algorithm so you have both MiewID and Hotspotter matches to check against.
Hello Anastasia,
Thank you very much for your efforts. I have gone through the entire Bulg import ‘minimum_import_2026_04’ and corrected the incorrect annotations. The result was better in some cases, but there were still errors where individuals could not be correctly assigned. Here is a link where I have manually assigned the number DD_RB_178 to the individual after checking all the hundreds of photos by hand. Wildbook
Afterwards, I uploaded a new dataset from the Roßthaler Bach (minimum_import_2026_05). The problem has cropped up again here: on the one hand, I have to correct almost all the annotations manually. After I’d fixed the errors and analysed the data against the database again, it once more suggested more individuals from the same survey date, rather than taking into account individuals from older Bulg imports. Here is an example:
thanks a lot
I suspect there may have been an unintentional change to the detector model during the Wildbook update that’s making it perform worse than before. It’s normal for it to mess up an annotation here and there, but this is excessive. The query annotations and the ones from many of the match candidates are still not correct as they are not including tails or heads. That will continue to have a downstream impact on your matches. I’m fixing a few as I find them and restarting matches, but this is such a broad issue that it’d be better to tackle it at the source, which is to troubleshoot the detector.
I hope to have an update for you soon. Thank you for flagging this.
