Active Part Segmentation

Annotations for large datasets is expensive
Human “corrects”/varifies the labeling of the automatic annotations
Re-train automated algorithm based on new annotations
Optimize for best use of annotater time per accurate labels