✳ BETWEEN QUESTIONS ML STUDY ATLAS

A QUESTION-LED LESSON

Object Detection: Where Is the Object?

Image classification can say “cat”. What if we also need to know where the cat is?

Labels and locations

An object detector predicts categories and locations, commonly as bounding boxes. One image can contain multiple objects, so a detector may return several predictions.

Compare boxes

Move the orange predicted box. The green box is the reference. Intersection over Union (IoU) is the overlap area divided by the combined area of both boxes. Exactly matching boxes have IoU 1; boxes with no overlap have IoU 0. The example uses equal heights, so comparing widths is enough. Move the predicted box away from the reference box. The overlap shrinks while the union grows; IoU summarizes their ratio.

INTERACTIVE EXPLORATION

Slide a predicted box

The reference interval is [0, 2]. Move a predicted interval of equal width along the horizontal axis; assume equal box heights.

Evaluation is more than one box

Detection metrics consider class, confidence, matching predictions to ground truth and IoU thresholds. Our experiment varies one predicted box in one dimension with identical height; it is a toy overlap calculation.

Check your understanding

What is IoU when two equal boxes coincide exactly?

1: their intersection and union are the same.

Keep exploring

What does image classification predict? ↗How is an image represented? ↗

Further reading

Source and further reading ↗