This module processes regions of interest (ROIs or vignettes) extracted by ZooProcess from an original image taken by the ZooScan instrument.

These ROIs should contains only one object for later classification. This module predicts whether an ROI contains multiple objects, so that it can then be further separated before their classification. The separation is done by another module called zooprocess_multiple_separator.

This application uses a MobileNet v3 classifier trained towards maximizing the recall of the `multiple` class. It returns the probability that the image contains multiple objects (and a natural threshold to predict it in the multiple class is therefore 0.5; but it could be lowered to further increase the recall of multiple objects).

This module has been developed through the iMagine project.