Image Segmentation

Semantic and instance segmentation, and the Dice metric.

20 minutesComputer vision

What you will learn

  • Telling the three types of vision tasks apart
  • Knowing the Dice metric

Segmentation classifies every pixel: semantic segmentation (road, sky, car) or instance segmentation (each car separately). It matters in medical images for outlining organs or lesions.

It is evaluated with Dice and IoU at the pixel level. Like detection, it is beyond the basic tools here, and we present it as a concept.

Implementation steps with the tool

  1. Compare: one image → a label (classification), boxes (detection), a mask (segmentation).

Quick review question

Which task do you need to compute the area of a tumour in a research medical image?

Tools

Related algorithms