Skip to content

✏️ The Google Quick Draw dataset ported to a MNIST-like 28x28 greyscale array dataset, useable by JavaScript machine-learning libraries

License

Notifications You must be signed in to change notification settings

wagenaartje/quickdraw.js

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

31 Commits
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Quick, draw! JS


This library allows you to import any of the millions of drawings from the Google Quick Draw dataset and to use them with JavaScript machine learning libraries like Synaptic and Neataptic. At this moment, this library is not supported by the browser due to size issues.

Example usage (using Neataptic)

const dataSet = quickDraw.set(100, ['car', 'airplane']);
const network = new neataptic.architect.Perceptron(dataSet.input, 20, dataSet.output);

network.train(dataSet.set, {
  iterations: 100,
  log: 1,
  rate: 0.1
});

This example will teach a neural network to distinguish if a drawing is either a car or a plane, using a dataset size of 100.

Getting started

Installing quickdraw.js is easy with npm, just run:

npm install quickdraw.js

Then, in your Node.js script, require the library:

var quickDraw = require('quickdraw.js');

Usage

Usage is fairly easy, but keep in mind that the Google Quick Draw dataset is a very large dataset. So before going mayhem on downloading, take into account that 100 samples are about 40kb (for 28x28).

With about 100000 samples for each of the 345 categories, that would take gigabytes of space.

quickDraw.set

This allows you to get a dataset that is compatible with Synaptic and Neataptic. Default usage:

quickDraw.set(size, categories)
  • size is the total amount of samples you want in your dataset. The amount of samples will be distributed among the categories the best way possible.
  • categories is optional. If none is given, samples of each category will be present. When given, it should be an array containing the names of the categories you want to include. See the full list here. All given categories should have the same sample dimensions!

Example:

var set = quickDraw.set(600, ['car', 'airplane', 'bicycle'])

This example will return a dataset with 600 samples, containing 200 samples from each of the 3 given categories.

The set function returns an object which is built up like this:

{
  set: [
    { input: [0, 0, 0, 0.418, 0...], output: [1, 0, 0]},
    { input: [0.156, 0, 0, 0.163, // one object for every sample
  ],
  output: 3, // amount of categories
  input: 784 // dimension of samples (28 x 28)
}

The .set property should be passed to Synaptic or Neataptic as the dataset. The input and output parameters are useful for setting up the input/output size of the neural network.

quickDraw.importAll async

By default, this library comes with 100 samples of each category. Each of these samples is a 784 (28x28 image) array containing values in the range of 0-1, where 1 indicates 'fully drawn'.

If you want to change the amount of samples per category, or the dimensions of the dataset, use this function the following way:

quickDraw.importAll(samplesPerCategory, dimension);

So for example, calling:

quickDraw.importAll(250, 64);

Would download, convert and save 250 samples per category as a 64x64 greyscale array.

quickDraw.import async

This is similar to quickDraw.importAll, but this function will just update the samples of óne category. The usage is fairly simple:

quickDraw.import(category, amount, size);
  • The category parameter is the string name of the category that you want to update. See the full list here.
  • The amount parameter is the amount of samples that should be downloaded for the given category.
  • The size parameter is optional, the default is 28 (28x28).

Example:

quickDraw.import('broccoli', 400);

Will download, convert and save 400 28x28 samples from the brocolli category.

quickDraw.checkSet

Checks the properties of a locally saved set. Basic usage:

quickDraw.checkSet(category);

So if you want to check the properties of the banana category, the code would be:

quickDraw.checkSet('banana');

The function returns an object containing the following properties:

  • size the amount of samples locally saved of the category
  • dimension the dimensions of the array (28 = 28x28)

Contributing

Feel free to contribute! Keep in mind that this project is fairly new, and I have some things in my head that I want to implement. So before you create a PR to work on a new feature, create an issue so we can discuss the feature first.

You can also participate in the discussions at the issues section, every opinion we can get is useful.

Further notices

This data made available by Google, Inc. under the Creative Commons Attribution 4.0 International license. https://creativecommons.org/licenses/by/4.0/


You made it all the way down! If you appreciate this repo and want to support the development of it, please consider donating 👍 Donate

About

✏️ The Google Quick Draw dataset ported to a MNIST-like 28x28 greyscale array dataset, useable by JavaScript machine-learning libraries

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published