This is an interactive, coding-free demo where you will train and test machine learning models on drawings that you draw!
For a taste of what's to come, draw a large, centered digit in the box below and see how a prebuilt model labels it.
Generate and label data to build a dataset.
The goal is to teach the neural network to distinguish between different categories of drawings. This could mean identifying a drawing of a '1' from that of a '0', the character 'A' from the character 'B', or a smiley face from a frowny face. I recommend starting by drawing 3 ones and 3 twos in the box below, labeling them as '1' and '2'
View and modify to your data set. Hover over an image to relabel it or delete it from the dataset.
A model is a function whose output depends on a set of weights and an input. In this case, the input is a picture and the output is a label. This training step finds weights that adjust the output to be the correct label for the pictures in your dataset. Hopefully, this same model will correctly label pictures that it hasn't seen.
Test how well the neural net has learned your dataset.
Select a model to test and draw a picture from one of your categories in the box below. As you draw, the selected model will label your drawing. See where the model succeeds and where it fails. If it is not performing up to your standards, add some more pictures, retrain it, and test it again.