Can you describe what the issue is you’re experiencing?**
We are having one of our seals, RHS03 completely dominating the match results, seals that do not look anywhere near him are not showing up in the suggestions and, maybe I’ve been lucky but in the past don’t remember multiple of the same individuals being suggested in match results.
We are also noticing incredible easy matches are being missed now, completely.
It is especially frustrating when we have uploaded RHS02 with his signature neck patch, and he did not even appear in the suggestions, but again RHS03 was at the top. Prior to this last update Sealbook seemed incredibly powerful and we are not sure what is happening at the moment.
I may not have an update for you until next week (and Monday is a federal holiday for us). Right now I’m not seeing your match results filtered by project, which I need to dig into first. I see the same results whether or not the project filter for Howe Sound is applied:
I have a few more questions while I dig in to this more.
For your first link (Wildbook) is the correct seal supposed to be RHS02?
If so, that could be just a case of a missed match since 1) Hotspotter also missed the match and 2) this is more of a front view than a right side view where the distinct cheek pattern isn’t as visible. You mentioned that this is happening more frequently overall, so I’m curious to see additional match page examples where the correct individual isn’t suggested.
As far as multiple candidates being of the same individual, that occurs more frequently with the Image score view instead of the Individual score view (the tabs below the Match Result heading). By default, we display individual scores, which computes one match score for every individual in the database. This is the aggregate of each image score for that individual. The image score computes the match score for every annotation in the database when compared to the query annotation.
The first photo was not supposed to be RHS02, I believe it could be a new individual it was just a second example of a strong bias for RHS03. I i can do some digging if it helps, but this is my first time updating our catalogue since returning from a leave, and I am the only one doing it - and the correct individual was not being recommended quite often- enough that I had started doing it by analog.
If it would help i can jump back and look through our catalogue for more examples
Gotcha. I’d say that match result is accurate then because of the way individual match scores and image match scores are filtered. I can see in your first screenshot that the image score is the highlighted tab which would make the likelihood of a single individual appear more than if you sorted them by individual scores.
The second link of RHS02 does appear to be a missed match as I do see other encounters of it that have a clear right viewpoint. Occasional missed matches are expected.
I don’t think these particular examples point to a trend, but if you do see more instances of known individuals not appearing as match suggestions, share them here so I can take a closer look at what could be behind it.
