Sharkbook MiewID results inconsistency

Support
,
1

Sharkbook - using the MiewID algorithm

As we know, MiewID has better results than previous Groth and i3S algorithms. So I have been running a review of every shark individual and re-testing the images using the MiewID algorithm to see if each individual shark is indeed an individual. So far, after re-revieweing over 70% of sharks I have found that 16% of sharks were in fact not unique afterall.

However, today I kept finding unusual results and think this is a bug within MiewID.

I have found situations where one image has been uploaded twice at different times. Image A is Image B. Yet when I run results on Image A it does not return Image B in the top results. But when I run results on Image B it finds Image A right away and returns it as its top result.

If Image A and Image B are exactly the same - why wouldnt both search results return each other as the best match? Why is it only occurring in one direction?

Here are some examples:
Example 1, where Image A does not return Image B in the results, but Image B returns it straight away:
Image A: Sharkbook: Wildbook for Sharks
Image B: Sharkbook: Wildbook for Sharks

Example 2, where Image A does not return Image B in the results, but Image B returns it straight away:
Image A: Sharkbook: Wildbook for Sharks
Image B: Sharkbook: Wildbook for Sharks

Example 3, where Image A does not return Image B in the results, but Image B returns it straight away:
Image A: Sharkbook: Wildbook for Sharks
Image B: Sharkbook: Wildbook for Sharks

This bug is truly puzzling me as I have found it frequently and it makes me fear that the algorithm is not reviewing all images or something. Please can you investigate and let me know what you find?

2

For this pair, Image A has a location ID of Wolf Rock and Image B doesn’t have a location ID. It only has a location description. Image A’s results will only look for other match candidates within Wolf Rock. Image B will look at the entire Sharkbook database for matches when no location ID is defined.

Ditto above. Image A has a location ID and Image B doesn’t.

Same explanation as above.

This isn’t a bug; matching is working as intended. If you update the location ID for all of the Image Bs and then re-run those matches, you should see improved match candiates.