What is the best way or standard practice for handling out of focus objects in an image, if you would prefer the model to ignore them? I’m talking about objects you would label if more in focus, but as they’re not, you’d rather skip them.
Excessively blurry objects won’t be fit for my purposes, so is it best to create another label for out of focus objects so that the model will differentiate and I can handle them in code? Or is it best to leave them and trust that they won’t muddy the model as you’ve not included them in your labeled objects?
I don’t think there’s a right answer here–it entirely depends on your use case. Do you want blurry objects to be identified as your target object? If not, then they should probably be left as part of the “background” class. If identifying “blurry” vs “not blurry” objects of the same class is important to your use case, then you should create two different classes. If you’re worried that the camera might not focus on the objects in time during inference (but still want to identify the object), you might want to lump “blurry” and “not blurry” in the same target object class.