Mask R-CNN

Introduction

Mask R-CNN is an algorithm that aims to resolve instance segmentation

Recall: Faster R-CNN

D-DL4CV-Lec15e-Faster_R-CNN

Implementation

In original faster R-CNN, for every region proposal, we have

  • Classification branch: predict object class
  • Bounding box regression branch: Revise the bounded box coordinates

Now, we add a new Mask Branch, which is a fully convolutional network that can classify each pixel of region proposal to object and background