What Wildbook are you working in? Internet of Turtles
Can you describe what the issue is you’re experiencing? During the Match Result process for an individual, we are trying to consider more options by increasing the “Number of Results.” However, the results did not changed from the previous options (we try to make it as 30 but keep showing 3 options only).
Could you please elaborate on how we can increase the number of results and whether the results are accurate enough with the updated algorithm (MiewID msv4_v3)?
My initial guess was that this match was run before IoT finished re-indexing annotations after its upgrade since your link shows “Against 3 candidates” on the match results. This would usually be resolved by re-sending this match to identification.
When I re-ran the match it increased to “Against 1501 candidates” but still only displays three turtles.
I’ll need a bit more time to look into this.
Thanks for your patience! This version of MiewID functions similarly to Hotspotter in that it will only display match candidates that have a match score greater than 0. If the match score is below zero, it won’t display any more candidates, no matter the number you change to at the top.
Would it be possible for us to retain access to the previous version of MiewID, at least alongside the new version?
Since the quality of our photos can sometimes be inconsistent, we still rely quite a bit on manually identifying individual turtles by eye. Having access to both versions would give us more flexibility and help us make sure we don’t miss potential matches, given that we can adjust the match cadidates. Thank you!
That’s a good question! Let me ask the devs when I meet with them today and I’ll get back to you.
Thanks for your patience. The MiewID behavior before the update would choose random turtles to display to meet the number of results requested, even when they were below the confidence threshold for what was considered a possible match. So if the top 12 results were all potential candidates based on the match scores, updating the number of results to 30 would have added 18 turtles that matched the location ID and taxonomy of the encounter, regardless of the matchability to the original turtle. While the update may look like you’re seeing worse performance, it’s actually reducing the number of false positives.
Instead, you may want to try the Visual Matcher option on the encounter page (the button is below the encounter image). This tool was created before we had computer vision available and will display turtle candidates based on location and species only, similar to what MiewID was doing before the update. Visual Matcher was originally used for mantas, but it was popular with researchers in helping find correct matches for other species in different Wildbooks, so we kept it in the update.
I see, understood. Thank you for the suggestion
1 Like
