good morning,
I have done some new entries. Sometimes they are only being matched by miewID and sometimes only by Hotspotter.
MiewID is absolutely not working for the Thornicroft giraffe yet, so when it is only matched with miewID this is not helpful.
Can you have a look at this?
thanks,
Frederike
Are you seeing inconsistent algorithms being applied for the same species? For example, Thornicroft giraffe photos sometimes display Hotspotter results + MiewID and other times just MiewID?
Yes that is correct. Sometimes only Hotspotter results, sometimes only Miew ID and sometimes both.
@Anastasia it seems like Giraffespotter isnt working… the whole day yesterday and now still not working?
Sorry for the late reply - I restarted the server this morning and it’s back online again!
Can you share links that show examples of the inconsistent results so I can research this?
only miewID:
Wildbook for Giraffe (giraffespotter.org)
Wildbook for Giraffe (giraffespotter.org)
Wildbook for Giraffe (giraffespotter.org)
Wildbook for Giraffe (giraffespotter.org)
only hotspotter: Wildbook for Giraffe (giraffespotter.org)
Wildbook for Giraffe (giraffespotter.org)
both:
Wildbook for Giraffe (giraffespotter.org)
Wildbook for Giraffe (giraffespotter.org)
These are great examples; thank you! We’ve verified that this is a bug and we’ll need more time to figure out why the algorithms are running inconsistently when both MiewID and Hotspotter should run together for Thornicroft’s giraffes.
Here’s the ticket link: Manually reporting an encounter doesn't consistently run default algorithms · Issue #487 · WildMeOrg/Wildbook · GitHub
In the meantime, I’m waiting to hear back from the machine learning team on the status of Thornicroft’s giraffes for MiewID.
Here’s the update on MiewID: We don’t have enough data to train it on Thornicofti yet. It primarily works best right now on reticulated giraffes.
You previously shared 10 individuals for the training efforts, but it’s not enough data for us to complete training it. We need at least 100 individuals for successful results.
LG-0001 - LG-0070 can be used for training MiewID already. I have an intern checking all the individuals on any mistakes. I will let you know when he is done with the first 100.
1 Like
Hi @Anastasia,
LG-0001 until LG-0100 can be used for Miew ID training.
Let me know if you need more,
thanks!
No, not yet. Any progress updates will appear in the ticket I linked, but I’ll update this post when it’s resolved.
Hey @FrederikeOtten
I have a quick update on MiewID: we’ve added the latest multi-species model to GiraffeSpotter (trained on 53 species) and are interested in hearing if you start to see improved match results with it.
Hi @Anastasia, it seems that at the moment the matching isn’t working. It doesn’t show any giraffes to compare.
hi, it does seem to work when a start a new match. MiewID is working a bit better, but Hotspotter is still much better.
1 Like
hi @Anastasia, any chance the above problem can be looked into? ‘match results’ don’t show matches, only when starting another match. thanks
Sorry, @FrederikeOtten! I think I misunderstood the problem when you mentioned it earlier.
Just to make sure we’re on the same page, submitting a new encounter doesn’t display match results after ID is run and it has to be re-run manually in order to work correctly. Is that right?
HI @Anastasia, my problem at the moment is that if you click on ‘match results’ you get no results:
Only if you click on ‘start another match’ you get results but obviously this takes much longer to get results.
I’ve spot checked your recent encounters and I’m seeing match results on the ones I’ve checked so far. Can you send me the link to some encounter examples to investigate? If you notice that they all come form the same bulk import, it may be a good idea to re-run ID on it because it could point to a WBIA hiccup that occurred when ID was first run.