lasagne-draw
lasagne-draw copied to clipboard
skaae
Implementation of the DRAW network in lasagne
DRAW implmentation [Gregor 2015]
Implementation of the draw network in Lasagne (https://github.com/benanne/Lasagne)
The Read/write operation is based on https://github.com/jbornschein/draw/tree/master/draw
This is work in progress and is poorly documented
Animation of the DRAW network reconstructing images. Modelling p(x).
.. image:: https://raw.githubusercontent.com/skaae/lasagne-draw/master/animaion.gif :alt: DRAW animation :width: 679 :height: 781 :align: center
Animation of reconstructing images modelling p(x,y). The two first rows generate 0's, next two rows 1's etc. The model is not fully converged
.. image:: https://raw.githubusercontent.com/skaae/lasagne-draw/master/animaion_cond.gif :alt: DRAW animation :width: 679 :height: 781 :align: center
Install
First install Lasagne and Theano.
git clone https://github.com/skaae/lasagne-draw.git
cd lasagne-draw
python setup.py develop
cd ..
python -c 'import deepmodels'
References
- Gregor, K., Danihelka, I., Graves, A., & Wierstra, D. (2015). DRAW: A Recurrent Neural Network For Image Generation. arXiv Preprint arXiv:1502.04623.
skaae