Link to Dataset: https://quickdraw.withgoogle.com/data

Link to Blog Post announcement of Sketch-RNN code: https://magenta.tensorflow.org/sketch_rnn

Link to Paper: https://arxiv.org/abs/1704.03477

Link to GitHub Repo of Sketch-RNN code: https://github.com/tensorflow/magenta/blob/master/magenta/mo...

Link to previous Hacker News discussion: https://news.ycombinator.com/item?id=14114706

I was wondering why almost all the drawing by people were "terrible"... and then I saw how they acquired them: https://quickdraw.withgoogle.com

Really hard to do with a mouse in the time allowed... but kind of fun.

The page works on our phone too, which is better.

I think this allows noise elimination by averaging though, it distills the key elements of a doodle of a bird, cat, car, etc.

In retrospect this was a genius way to get training data :)

Can I use this to work out if something is a hot dog? I've got a killer app idea...

It's​ not that precise, it will only determine whether or not a drawing contains a sandwich.

Better if they release the trained model.

the jupyter notebook will download a bunch of pre-trained tensorflow lstm models into /tmp/sketch_rnn