Euro coin sorter - How does the model understand the different natures of the same object?


I am new to the forum and this is my first intervention.

Following the first exercises of the course, I have set out to identify the value of existing euro coins.

I have build my model from resnet34, using 8 classes for each coin value.

In my original idea I am not discriminating both sides of the coin, so I am associating heads and tails to the same label with the value of the coin.

Here are my first concerns about how the system learns:

Will the system associate the characteristics of both faces to a single class or will it only try to find the common features between both faces? (M∩N or M∪N or ???)

In the case that the system tries to find the common features between both faces, what happens if these two faces have nothing in common? In this case, it is clear that the best thing would be to make two different classes for each currency, for example: face_1 cent and cross_1 cent. But the problem would be complicated if we consider that in the case of euros, each country has its own face. So the ideal would be to find a system that would associate different natures of the same class.

I will developpe this forum thread, as a register of what I am descovering and to expose the different doubts which will arise in the process. I would greatly appreciate if anyone could help narrow the path.

Greetings, in advance thank you for any clarification or additional information.