Quick Scoop

Semantic segmentation is a computer vision task where a model assigns a class label to every pixel in an image, so it can tell not just what is present, but where each thing is.

In plain English

If image classification says “there is a car in this photo,” semantic segmentation draws out the car area pixel by pixel and labels those pixels as “car”. Unlike instance segmentation, it does not separate multiple objects of the same class into different individuals; it groups them by category instead.

Why it matters

This kind of detailed labeling is useful when boundaries and shapes are important, such as in self-driving cars, medical imaging, agriculture, and industrial inspection. It helps systems build a dense “map” of an image rather than just giving a single prediction for the whole picture.

Example

In a street photo, semantic segmentation might label:

  • road pixels as “road.”
  • people pixels as “person.”
  • sky pixels as “sky.”
  • cars pixels as “car.”

Tiny takeaway

Semantic segmentation is basically pixel-level understanding of an image.

TL;DR: Semantic segmentation labels each pixel in an image so a model can identify object categories and their exact locations.