Since the data is mostly round petri dishes, the network should not be using the shape of the objects for predictions, but rather the hyperspectral data. It makes sense to leverage augmentations as this might help the network to focus on the hyperspectral "texture" rather than the shape.
Since the data is mostly round petri dishes, the network should not be using the shape of the objects for predictions, but rather the hyperspectral data. It makes sense to leverage augmentations as this might help the network to focus on the hyperspectral "texture" rather than the shape.