Mask R-CNN

← Back to the article

Mask R-CNN finds objects and colors the exact pixels that belong to each object.

What it does

Mask R-CNN finds objects and colors the exact pixels that belong to each object.

Segmentation: Outlining an object • Instance: A single specific object

How it works

  • Find an object in the image.
  • Draw a box around it.
  • Make a mask that covers the object shape.

Uses and applications

  • A medical tool can outline a tumor area for a doctor to inspect.
  • A photo editor can select one person from the background.
  • A robot can understand the exact shape of an object it should pick up.

In short

It finds precise boundaries, such as a tumor in a brain scan.

Related methods

  • Faster R-CNN: The detector that Mask R-CNN builds on.
  • U-Net: Also produces pixel-level masks, without separating object instances.

Back to Computer vision · All topics

The interactive demonstration and the Python walkthrough for this topic need JavaScript.