What operating system were you using? (eg. MacOS 10.15.3) MacOS 12.7.2 and other
What web browser were you using? (eg. Chrome 79) 127.0.6533.120
What is your role on the site? (admin, researcher, etc) admin
What happened? Hi, in the last year/past 6 months or so I have seen a marked increase in the ability of Sharkbook of detecting the nursehound sharks, even from partial or poorly lit photos.
However in the last couple of months or so I have seen a massive decrease in efficiency from the system, together with a few glitches which you quickly sorted out (just mentioning as they may be part of the reason why?). Now it presents only a couple of candidates, none of which are a good match, whereas on visual examination I can ID individuals (which are in the system, so they should have been ID’d easily)
I am not complaining because you are doing an amazing job for us, just checking in if there’s anything that can be done?
What did you expect to happen? Better matching and more candidates that I can visually check!
What are some steps we could take to reproduce the issue?
For example:
Whereas this was a good match, but I found it strange that it presents only 6 candidates: Sharkbook: Wildbook for Sharks
Many thanks for all your invaluable help!
PS - also: any chance we can start an ID programme for Myliobatis aquila, the common eagle ray? They are brownish but some have white or black spots, and some have cuts on their fins or tail which can be cut off. I guess it would work for common stingrays Dasyatis pastinaca, which I am starting to have a few as well - they are greyish but similarly with whitish markings.
Many thanks! Eleonora
If this is a bulk import report, send the spreadsheet to services@wildme.org with the email subject line matching your bug report
Do you have examples of previous match results that had a bigger list of match candidates? Looking at your examples above, there are only about 300 encounters of this species in that location. I’m curious if the older match results that had more candidates to view may have been from a different location ID.
Additionally, Hotspotter will only display match candidates that have a match score greater than 0, even if you ask for it to display more than the default number of candidates.
We do have spotted eagle rays on MantaMatcher, but not common eagle rays or common stingrays. Have you been able to reliably ID individuals from these spots?
Ciao Anastasia, 99% of pictures are indeed from the same location BSC, and 300 pictures seems about right for all the pix we have.
Roughly speaking, last year I would set the n of candidates to 48, visually inspect, and the right match would pop up way down the list. A few months ago, instead, the maching would provide only a few candidates, but bang on - the perfect match would pop up in the first 3 options. Now, alas, very few candidates and it would not recognise a shark that was positively ID several times
The good news is that just now I was just able to get a positive match Sharkbook: Wildbook for Sharks , but my worry is that it said
Matched…against 4 candidates
Surely there should be more than 4 candidates?
On another note, moving the cursor over the candidates score now won’t result in the corrisponding new image to check.
Thanks for checking out!
As for stingrays and eagle rays, I am actually doing the ID now. It works fine within this season, and will start looking at past years to find some matches…
We have been refining the algorithms in Sharkbook over the last few months and I confirmed that as the algorithms improve, it’s normal to see fewer match candidates displayed. This is because as it gets better at identifying individuals, it will present fewer matches that are not as likely to be your shark. Additionally, as individuals are assigned, the match candidate pool tends to be reduced because a successfully identified individual will have its own individual score (you can see this button on the top of the match results page). This computes one match score for every individual in the database and is an aggregate of each image score for that individual.
Yes, this is weird. I’ll need to check with a teammate who’s out of the office right now, because my expectation is that the number should be higher than 4 here.
You have to click on each result now to display it instead of hovering over it. This change was part of the Wildbook 10.1.0 release.
MiewID is coming to MantaMatcher in the future. Right now all mantas there are spot-mapped from photos of their undersides to ID them. I’ll send you an email so we can talk more about the additional species support with the rest of the team.
OK, I have an update about the 4 match candidates. A recent upgrade classed these internally as whale shark annotations instead of nursehound sharks. We’ve fixed this, but you may want to re-run ID on the encounters that only compared against a handful of candidates.
I have to agree that I am experiencing the same within the Carcharias taurus species. I never used to rely much on i3S algorithm, but recently the modified groth algorithm has been really poor. The other day I reviewed ~150 sharks before I got to the correct one.
On this example, the only match it found was the duplicate image. I looked through all 15 of the modified groth matches, and I looked through all 91 of the i3S matches, and the only match it found was the duplicate image. It should have matched to the image above.
Unlikely, since nursehound sharks don’t use spot-mapping. It looks like Carcharias taurus is included in MiewID for Wildbook 10.3.0, which hasn’t rolled out to Sharkbook yet. We’re still in the process of updating all of the wildbooks to 10.3.0.
I’ll need to check with a teammate who’s out of the office right now, because my expectation is that the number should be higher than 4 here.