Also note This photo isn’t a great one for the algorithm due to the remoras and shadow, but I was running the search to show someone how it works, and I was a bit surprised to see the wrong species pop up in the results.
If this is a bulk import report, send the spreadsheet to services@wildme.org with the email subject line matching your bug report
It was with the old algorithm. The newer one was still loading and I was trying to demonstrate for someone. (The newer ones are showing no matchable detection). It’s not a good photo for the old algorithm, I was just using it as an example of what the results look like.
We are not putting any more effort into the older algorithm, in preference for HotSpotter and PIE, both of which we have a long-term interest in maintaining, and both are only run within a single species. Here is that same Encounter run with the newer algorithms:
When I went to the Encounter, I did see that a matchable pattern had been extracted (it gets created whenever you do the crop/rotate for the old algorithm as well and gets the “CR Image” keyword but no pattern match had been kicked off. The CR Image photo does look different different than the regular, green highlighted detections, as shown below.
We are approaching full automation with mantas, working with Olga Moskvyak on an “orientation network” to standardize detected manta pose, which will hopefully replace the need for crop and rotate entirely.
