Hi, in Sharkbook we still have to rely heavily on the Modified Groth and i3S algorithms to identify sharks.
Recently however, spot maps are rarely coming back with more than 5 potential matches (it normally returns around 50 matches).
Could you double check if its working properly?
Here are some examples:
Thanks
Sarah
I just re-uploaded a shark encounter and did a spot map to demonstrate the issue.
In the previous spot map, it produced a number of results.
In the new spot map, no results came up for either Modified Groth or i3S:
Check the spot map results for these two encounters:
Thanks for your patience! Thankfully, this wasn’t a bug. We re-ran both links and the one with the missing spot map shows results and that 6930 patterns were considered. The other links results also look good and I also had Jason verify that it’s normal.
If you see questionable spot map matches again, try waiting a little bit and then re-running the match in case the questionable result was a hiccup in the identification process.
Thank you Anastasia!
I am really looking forward to when I can completely rely on MiewID. I am still finding it bias (eg if the shark is pregnant, it returns only pregnant sharks. Or if the shark is front facing, it is bias towards showing other sharks front facing).
The MiewID algorithm already appears to be improving though. Thank you again for all of that!
Thanks for this feedback! I’ve shared it with our machine learning team.
Here’s the response I got back from the team:
It makes sense with the way our training subset is. Also, we have just a few aquatic species being matched fullbody, which I think makes the algo go for the ‘easy way out’. I will keep this in mind with future retraining and see if I can compensate with a few tricks.!
Also, if she is willing to contribute the data she has reviewed - super welcome!
I will definitely add feedback then if it will help! Thank you - do you prefer in here, or via email?
Email is perfect!