Object Detection

The concept of detection, the IoU metric and the limits of the current tools.

20 minutesComputer vision

What you will learn

  • Telling classification and detection apart
  • Computing the IoU of two boxes

Detection locates each object with a bounding box and its class, and an image may contain many objects. It is evaluated with IoU (the overlap of the predicted box with the true one) and mAP.

The Academy's five no-code tools do not offer training an object detector directly in their basic versions, so we present the concept and say so plainly. Other specialised platforms exist, but we have not added them before verifying them.

Implementation steps with the tool

  1. Compute IoU by hand: two boxes that overlap by half of their area.

Quick review question

A predicted box covers only half of the object. Is the IoU high?

Tools

Related algorithms