Skip to content
BENIBIX Learn

Depth Estimation

Also known as Monocular Depth Estimation, Depth Map, Depth Prediction

Working out how far away each part of a photo is. It makes a depth map.

Think of it like

It is like guessing how far away a tree is just by looking. Small and hazy means far.

Example

Lucy's phone blurs the background in portrait photos. A depth model guesses which pixels are her and which are the wall behind her.

Why it matters

Robots need distance, not just a picture, to grab things. A depth map gives it from a cheap camera.

Definition

Depth estimation works out how far away each part of a picture is. The answer is a depth map, an image where each pixel holds a distance. With two cameras side by side, the model compares the views, as your two eyes do. With one camera, a model learns clues like size, shade and blur. One camera is cheap, but its guesses are less exact than lidar.
On the right, one photo becomes a depth map. Each shade stands for a distance.

Common mistake

Trusting a single-camera depth map as exact. It is a learned guess, and glass, mirrors and plain walls can fool it.

Key takeaways

  1. 1
    Distance per pixel. The output is a depth map.
  2. 2
    Two eyes or one. Compare two views, or learn the clues.
  3. 3
    A guess, not a tape. Lidar measures more exactly.

Compare