I had around 160 photos of sea turtles to match, and the majority of them have been matched with MiewID Deep Learning Match, and only 2 with Hotspotter.
My first question is: what is the difference between these two algorithms?
My second question concerns the scale of the match results with MDLM and HS? Because most of the match results ranged between 0 and 0.8 something. Does that mean that the max is 1 (as in 100% match results)?
Also I noticed that the scale of the match results with Hotspotter seems different. How can I compare with the MDLM results?
Hotspotter is an older algorithm that checks for patterns (hot spots) in an image. When determining match candidates, it will only display matches with a match score higher than 0. This is why sometimes you will see less than 12 candidates shown in your match results.
MiewID is a newer algorithm and uses deep learning to identify individuals and is trained on multiple species. MiewID has consistently shown a significant jump in performance and an advantage in making positive identification with real-world photo conditions compared to older algorithms.
This first set of numbers in the match results is the ranking. The following set of numbers are match scores. The higher the score, the higher the likelihood of a match. What the score represents is slightly different for each algorithm and will vary from one trained instance of the algorithm to another trained instance. We recommend focusing on the ranking instead of the match score.
Ok thanx! I asked about the score, because we are trying to compare the score between our platform and IOT. And on our platform, the scores are displayed as percentage. So I wanted to see how we could compare it.
Another question: is it possible to choose from IOT to use MiewID instead of HS? Or does it just choose itself?
To follow up on this, we’re actively working on better explainability on the match score page so that users have more information as to what the rankings, scores, and inspection results represent. You may have seen Community posts from our collaborators at CMU as part of this effort.
You can when you select “Start another match” from the menu on the image form a encounter page. You’ll see checkboxes next to the default algorithm for a species and you can uncheck the one next to Hotspotter.
In general though, there isn’t a way to pre-select this before the first attempt to run ID in either bulk import or the manual upload page. We have default algorithms configured for each species that will run automatically when encounters are uploaded. That said, once we have enough feedback about MiewID performance in IoT, Hotspotter will be disabled.