Give the picture a small address

All feature responses double

Botie has two image feature maps and wants one compact descriptor. A uniformly larger response should not by itself make a picture a stronger match.

BOTIE

Every map entry is twice the first case's entry. What should happen to the final normalised descriptor?

The inputs we are comparing

Flattened feature maps

Example 1

1	2	3	2
0	1	0	1

Example 2

2	4	6	4
0	2	0	2

Result for example 1

00.9701
10.2425

Choose an explanation to test

Every coordinate doublesIt remains the same

One image feature never responds

This check uses the pieces from “Summarise each feature map, then remove overall magnitude”.

BOTIE

Average each flattened feature map, then normalise the resulting nonzero descriptor to unit Euclidean length. Return one coordinate per feature map, including the inactive map.

Flattened feature maps: 3 3
4 4
0 0

What result should your machine produce for this case?

Use a number, text, or an array matching the requested result. For a matrix, put each row in brackets. Saving your prediction runs your current construction on these inputs and checks the earlier examples too.

Build a construction in the workspace above first. This case will test those pieces.

Which spatial arrangements could change without changing either feature-map mean?

Your notes for later. These stay with this journey.

A compact image descriptor preserves selected feature statistics and discards substantial image detail.

This is a small construction problem. The lesson explains the assumptions and the parts of the full method that this example leaves out.

Work through the Python challenges →

Continue the courseLearning the Metric Itself →

← U-Net