#mask-r-cnn

Question 1: Single choice

A colleague writes: "Segmentation predicts masks and does not tell objects of the same class apart; typical architectures are Mask R-CNN and U-Net." What is wrong with this claim?

Neither model predicts pixel masks; both of them output only bounding boxes1
Mask R-CNN predicts one mask per detected object, so it does tell instances apart2
Nothing; both models output a single class map and never separate object instances3
U-Net is the model that separates instances, while Mask R-CNN only labels pixels by class4

Timed test