Advanced Deep Learning with Python

Instance segmentation with Mask R-CNN

Mask R-CNN (https://arxiv.org/abs/1703.06870) is an extension of Faster R-CNN for instance segmentation. Faster R-CNN has two outputs for each candidate object: bounding box parameters and class labels. In addition to these, Mask R-CNN adds a third output—an FCN that produces a binary segmentation mask for each RoI. The following diagram shows the structure of Mask R-CNN:

Mask R-CNN

The RPN produces anchors in five scales and three aspect ratios. The segmentation and classification paths both use the RoI predictions of the RPN, but otherwise are independent of each other. The segmentation path produces ...