What Wildbook are you working in? Sharkbook
What is the entire URL out of the browser, exactly where the error occurred? Sharkbook: Wildbook for Sharks
Can you describe what the issue is you’re experiencing? Algorithm not configured using spot mapping
Can you provide steps on how to reproduce what you’re experiencing?
If this is a bulk import report, send the spreadsheet to services@wildme.org with the email subject line matching your bug report
Hi @drobinson
Spot mapping predates computer vision, so newer species in Sharkbook aren’t configured for it.
You’ll need to manually annotate for this species and then start a match from the annotation. Here’s the announcement we made about it: Wobbegongs in Sharkbook
Hey Anastasia,
We still use spot mapping, especially for leopard sharks, where the annotation doesn’t find a match. A lot of the time, when we use the spot mapping after the annotation has failed to find a match, the spot mapping finds the shark, such as the example below:
With Wobbies, I think the spot mapping will be even more important as they are so cryptic, a lot of the time the algorithm can’t delineate the shark, and so we would like th option to use spot mapping to choose the area for the algorithm to use. Would this be possible? I can send lots of examples where spot mapping has worked, and the annotation has failed.
That is, unless you can suggest better ways for us to identify wobbies; completely open to suggestion and what is the best way to do so.
Thank you
David
It’s not likely as our goal is to move away from needing to spot map altogether. Our dev team is only made up of a handful of people, so we do need to prioritize efforts toward long-term impact across Wildbooks.
Jason previously mentioned that once there are 1500-2000 photos of wobbegongs in Sharkbook, it would be sufficient to train the AI detector and replace manual annotation (and in turn, retrain the identification algorithm).
Do you have examples of wobbegong matches where you know who the correct individual is but they weren’t suggested as a match candidate?
Quick update: it looks like there aren’t very many annotated wobbegongs in Sharkbook yet. In order for them to appear as match candidates, they need to be annotated first. Once the existing encounters have been annotated, you should see improvements in match candidates as it’ll have more annotations to check against.
Hi Anastasia,
Thanks for your reply.
Spot mapping at the moment seems to be far more accurate than the annotations, so with leopard sharks, it would not be in the best interests of ID to move away from that at the moment until we figure out why.
Did you look at the example given and see why the annotation doesn’t pull up a match, but spot mapping does? It would be great to know if spot mapping is going to be removed in the future, as this happens with the majority of leopard sharks and I also see this with whale sharks too.
If we can work this out with leopards, this may solve the wobbie issue, too.
Thanks
David
We agree; that’s why spot mapping didn’t go away when MiewID was introduced like we originally planned. With additional training, we hope that MiewID will eventually outperform spot mapping. Until that happens, it will still be available for the species that still use it.
I did and there wasn’t a spot map result available for this encounter so Jason and I created one on Friday. Are you sure this was the encounter you meant to post as an example? LSBB-678 doesn’t show up as a match candidate in either the Modified Groth or I3S spot mapped results.
Hey Anastasia!
So I don’t run Modified Groth or i3s. I use the spot mapping function to create the spot pattern area (between the three dots) and then go back into the encounter and run a match using Miew ID and Hotspotter. This works well and pulls up many more matches than the annotations. Please see screenshots below. If you click on match results on both images, you can see the second (without annotation) matches to the correct shark, whilst the annotation match does not. The images below to show the matches. This is the function I want on wobbegongs as they are even more camouflaged; I don’t need i3s or MG…
Here’s another one from today:
Ah, interesting! You’ve essentially been doing a hybrid of spot mapping and MiewID. This specific workflow process is only meant to be used with whale sharks.
In general, when we refer to spot mapping, it isn’t just the act of placing the spots, but also using Modified Groth and I3S to interpret those spot maps and suggest matches. The exception to this is for whale sharks, where the spot mapped feature regions are needed in order to start matches from them using MiewID. Officially, only broadnose sevengill sharks and whale sharks still require the spot mapped feature regions in Sharkbook.
I can’t speak to the “off label” performance in species it wasn’t set up for, but I’m tagging @jason for visibility.
Hi Anastasia,
It was Jason who trained me in this method for leopard sharks, that’s how I knew what to do :o)
If we could do the hybrid with Wobbies, that would be great and definitely increase ID accuracy.
Do you know why the hybrid method works better than the annotation?
Thanks
David
I’ve emailed you with details for setting up a meeting since this seems like it would be better worked out in real time.
