This seems to be based on Google's QuickDraw datasets. 50 million samples are available in an open dataset,<p><a href="https://github.com/googlecreativelab/quickdraw-dataset" rel="nofollow">https://github.com/googlecreativelab/quickdraw-dataset</a>