Skip to content
BENIBIX Learn

Image Segmentation

Also known as Semantic Segmentation, Instance Segmentation, Pixel Segmentation

AI gives each dot in a photo a label. Then you see the true shape of each thing.

Think of it like

It is like a coloring book. You color every part of the page by what it shows.

Example

A photo app cuts Lucy out of a selfie. It marks each pixel as Lucy or not Lucy, so the cut follows her hair.

Why it matters

A box is rough. When the exact edge matters, as in a scan or a photo edit, you need each pixel.

Definition

Image segmentation is when an AI model gives each pixel in a picture a label. You get the exact shape of each thing, not just a box. One kind colors all pixels of one sort the same, so all cars are one color. Another kind also tells each car apart. It is used where the exact shape matters, such as a tumor in a scan, or cutting a person out of a photo.
Every pixel colored by what it shows: road, cars and people.

Common mistake

Using it when a box would do. Segmentation needs far more labeled data and computing power.

Key takeaways

  1. 1
    Every pixel. Gets a label.
  2. 2
    Exact shape. Not a box.
  3. 3
    Semantic or instance. Kinds, or each one.

Compare