Module 22

Networks that understand images

image, convolution, pooling, learnable filters

When you finish this module you can say:

I can treat an image as a table of numbers, write convolution and pooling by hand, and explain why a network learns its own filters.

By the end of this module you will be able to write convolution and pooling without a library and build the pipeline of an image classifier yourself.

Lessons

6 lessons · ≈ 62 min

1. An image is a table of numbers9 min
2. Convolution: slide a kernel over the image10 min
3. Padding, stride and pooling10 min
4. Classifying with features11 min
5. Filters are learnable10 min
6. Project: your own image pipeline and Colab12 min · project

Practice exercises

Short writing exercises, independent of the lessons: Go to the exercises

1. A function that shuffles columns
2. Pooling that swallows a shift
3. Learn a single weight