Google are a crazy company that do crazy things, but their Quickdraw AI game is a incredibly friendly way to show off the power of AI and machine learning to everday people, while getting developers excited about the possibilities.
The project comes was developed by Google staff, including Jonas Jonjegan (web developer) and Henry Rowley (machine learning researcher) who explain the game works by interpreting what you draw using machine learning. As you add detail to the picture, the algorithm’s confidence in what you have drawn grows by analysing not just what you’ve drawn, but in what order and how you’ve added strokes to the image. This data is compared to a massive dataset and analysed for how much it matches your drawing. The impressive part is that its being done live, not a result returned after hours of crunching.
The project launched back in 2016 and since then people have drawn a massive 1 billion doodles, including thousands of dogs. Those paying attention will notice the URL structure and see it ends in /data/dog which means you can see basically any drawing by exchanging your word of choice at the end of the address. Try, cat, fish, bird, house for example, or read the full list here.
For those developer types who want to go deeper, make sure you check out the GitHub project site. This contains a dataset of 50 million (while not the billion, its still pretty massive), across 345 categories and features the raw moderated (adult categories not allowed), the preprocessed dataset and a list of the projects already using the dataset.
Thankfully Google has licenced the works under the Creative Commons Attribution 4.0 International license. This is just one in a series of AI experiments from Google, the rest can be found at experiments.withgoogle.com/ai
Creative and artistic projects
- Letter collages by Deborah Schmidt
- Face tracking experiment by Neil Mendoza
- Faces of Humanity by Tortue
- Infinite QuickDraw by kynd.info
- How do you draw a circle? by Quartz
- Forma Fluens by Mauro Martino, Hendrik Strobelt and Owen Cornec
- How Long Does it Take to (Quick) Draw a Dog? by Jim Vallandingham
- Finding bad flamingo drawings with recurrent neural networks by Colin Morris
- Facets Dive x Quick, Draw! by People + AI Research Initiative (PAIR), Google
- Exploring and Visualizing an Open Global Dataset by Google Research
Code and tools
- TensorFlow tutorial for drawing classification
- Quick, Draw for Processing by Cody Ben Lewis
- Quick, Draw! prediction model by Keisuke Irie
- Random sample tool by Learning statistics is awesome
- SVG rendering in d3.js example by Ian Johnson (read more about the process here)
- Sketch-RNN Classification by Payal Bajaj
- quickdraw.js by Thomas Wagenaar
For the best explanation on how it works, check out the explainer video below, then go to https://experiments.withgoogle.com/ai/quick-draw and try it for yourself.