Hi
I see two possible differences between eMiLios choice and the others.
At first it seems that MichaeL loves a huge, blue sky.
The second is, that on the first photo is a complete bench, in the others only the plaque.
Maybe aMaLia have a problem to proof it automatically.
The biggest problem with machine learning is that it’s only as good as the data that are inputted. I don’t know how they create their models, but if what’s categorized as an acceptable, POI is based on results from community reviewers we know it’s going to be a mess.
I think there needs to be enough (whatever that may be) of a similarity between the 2 photos for Emily to agree that on the balance of probability it is the same object.
Emily will always be learning.
So the first one is a bench replacing a bench that’s a fair bet
2 and 3 will probably look like very different objects so computer/Emily says no.
4 is almost discernible as a bench but not quite……I think this where Emily may overtime and feedback learn to change the likelihood that it is the same……but my understanding could be very wobbly.
I think you are right that it helps for us to understand exactly what is happening so we can adjust our expectations and also our approach.