Classify your Doodles with Code

Challenges

Table of Contents

Challenges

Experiment with different drawings to see if you can get some understanding of how the neural network is making its decision. Try out the challenges below.

Try drawing things that are half one class and half the other (e.g. a poodle scooter!). How does the neural network cope with this? Here is my 'scooter-poodle':

See if you can find what "features" (i.e. parts of the drawing) are most likely to cause one classification or another. E.g. what are the fewest drawing strokes you can make to get a poodle classification? Here is my simplest scooter (I wonder how big I need to make it to turn it into an avocado!):

Create some "abstract" drawings that the neural network recognises. Here is my abstract poodle:

How can you extend the program to classify more classes of doodles?
How can you improve the accuracy?
Create a game which challenges the player to draw pictures that the neural network can recognise.

More Fun

Have a go at this fun game that uses the same techniques as we used:

Quick Draw

< Prev Page Next Page >Back to resources page

Table of Contents