High number of false negatives on IoT with MiewID

Hello!

What Wildbook are you working in?
Internet of Turtles

What is the entire URL out of the browser, exactly where the error occurred?
After the implementation of MiewID, the ORP team has recorded a high number of false negatives, i.e. encounters of well known turtles which did not come up with a successful match. I have a list of encounters compiled at this point from various locations, but just some examples below:

  1. GR1268 from Maldives Laamu, 141 reference encounters available, no match found here: Internet of Turtles

  2. GR824 from Maldives Laamu, 82 reference encounters available, no match found here: Internet of Turtles

  3. HK5759 from Maldives Baa, 10 reference encounters available, no match found here: Internet of Turtles (Note that encounters of the same individual uploaded later did result in better matches)

  4. HK5609 from Maldives Noonu, 4 reference encounters available, no match found here: Internet of Turtles When running HotSpotter, a successful match was found.

  5. GM056 from Oman, 29 no match found here for the left side, right side match spot 16: Internet of Turtles This seems to be a common theme for Oman - one side no match found, for the other side a match is found in spot 15-16 - maybe be

Please let me know if you need more examples.

Can you describe what the issue is you’re experiencing?
No matches found under the first 24 suggested matches for well known turtles when using MiewID, while HotSpotter seems more successful in finding matches, ranking the correct individual in the first spot.

We see a preference in matching with recently uploaded encounters, those seem to be listed higher in the suggested match order. Not sure if image quality of the image being checked does play a role here?

Mainly, this issue renders the matching algorithm more or less useless for our purposes at the moment, as too many turtles go undetected with MiewID, and Hotspotter is too slow as you know. This very much limits our option to run meaningful matches against multiple locations (i.e., atolls in the Maldives for example), which we always considered a very useful feature of the platform.

Can you provide steps on how to reproduce what you’re experiencing?
Check the currently found matches on the encounters above. Rerun the matches, if a more recent encounter for the individual has been uploaded by now, the match might be more successful in finding the correct individual.

Please let me know if I can provide any more details!

Best,
Steph

Tagging @jason for visibility

Hi @ORP-StephanieK

Thank you for the feedback! We have a new multi-species model (for 54 species) with better accuracy across different species including sea turtles. We can upgrade IoT to the latest model probably by Monday, if not sooner.

Hi @ORP-StephanieK

The new MiewID model was deployed to IoT last night. We tested one of your example match results from above and see the correct turtle in the top spot now.

If you try re-runnning your previous matches, you should see improved results by comparison.

Thanks again for calling attention to this!

Thank you, we will be testing it!

1 Like