Thanks.
No, i think i did things as described in the wiki.

Yes, image size is what i also suspect, but as i describe above, there are only down to 2 images of tiny size…
So, what should be a non-problematic size and amount?
Is it normal to just fail with the described error, or should i not see a more specific error if it is related to the data size?

Thank You.

I found this Comment on a Github Issue, pointing out that the “Object Recognition” feature might be what i need to try.

But from the Wiki, i understand that only the 20 pre-trained objects can be recognized. So, that's a dead end too.

But wait, reading through the docs again, it seems “Object Classification” is exactly what i need.

Will have to try that.
And i also found this Forum Thread

which shows distinguishing a face wearing a mask / no mask using Object Classification. So that's very close to detecting a smile.

Also, i have the Huskylens PRO, and in it's WIKI, an example using Mind+ is shown, that looks to me as if new objects CAN be learned this way.
I see that training images can also be used from the PC.

So, this is something to look at too.

Thank You.

And if I understand correctly, custom models are also not possible?

Huskylens is not supported by tools like Edge Impulse,  right?