Pytorch-Sketch-RNN

A pytorch module for A Neural Representation of Sketch Drawings.

This python module uses the pytorch implementation of SketchRNN made by Alexis David Jacq (source repo). Which, in turn, is an adaption of the original Tensorflow implementation by David Ha and Douglas Eck.

Changes from Jacq's work:

• Python module: use import sketch_rnn
• Picture drawing and hparams as saveable/loadable attribute
• HParams as separate class, that is saved/loaded with the model.

---

In order to draw other things than cats, you will find more drawing data here: https://github.com/googlecreativelab/quickdraw-dataset

epoch 1900:

epoch 2400:

epoch 3400

Default hyperparameters for training has been found here: https://github.com/tensorflow/magenta/blob/master/magenta/models/sketch_rnn/README.md