Mask R-CNN (2017)

Faster R-CNN + Mask Branch

Mask R-CNN is an extension of Faster R-CNN, with instance segmentation support. It adds the mask branch to perform instance segmentation, which runs parallel to the existing branches for classification and bounding box regression. It’s part of Facebook’s Detectron suite, and the source code is available here.

This article explains how the mask branch works. If you are curious about the other two branches, please take a look at the Faster R-CNN article for details.

Instance Segmentation

Object detection predicts object locations (bounding boxes) and their categories (classes) in an image. On top of that, instance segmentation predicts a binary mask for each object (instance).

The below images show instance segmentation examples from the Mask R-CNN paper. Each object has a segmentation mask (within the bounding box) to identify the pixels occupied by the object.

Figure 2 of the paper

Mask R-CNN is an extension of Faster R-CNN, which is an object detection model with two branches: one for bounding box regression and the other for classification. These branches operate on RoI (Region of…