sketch_rnn_pytorch

Implementation of RNN-VAE following the work of Sketch-RNN using pytorch. https://magenta.tensorflow.org/sketch_rnn

Implementation details

The loss consists three parts.

1. Mean square error of the decoder network
2. Cross entropy error of the stroke type
3. KL-divergence loss

Dataset:

https://console.cloud.google.com/storage/browser/quickdraw_dataset/sketchrnn/?pli=1 For cat: https://storage.cloud.google.com/quickdraw_dataset/sketchrnn/cat.npz