A shape recogniser with features and a tiny CNN

The problem

Recognise simple shapes (bars, crosses, squares, triangles and circles) drawn at different positions and sizes in small images, first with hand-made features and nearest neighbours, then with a small convolutional network.

Requirements

Technologies

Step by step

  1. Write the shape generator and look at a sample of each class as text.
  2. Build the raw-pixel baseline and record its accuracy.
  3. Add gradient-orientation features and compare the accuracy with the baseline.
  4. Build the CNN forward pass and check every output shape by hand.
  5. Train the CNN (or its final layer) with gradient descent and compare all three methods.
  6. Study the misclassified images and explain what each method gets wrong.

Expected result

Three working classifiers for the same task, with a comparison that shows why features and convolutions beat raw pixels.

How it is evaluated

Ways to improve it