In [1]:

    

# Load graph and restore weights

metagraph_fp = './train/infer/infer.meta'
ckpt_fp = './train/model.ckpt-6532'

import tensorflow as tf

tf.reset_default_graph()
graph = tf.get_default_graph()

saver = tf.train.import_meta_graph(metagraph_fp)

sess = tf.InteractiveSession()

saver.restore(sess, ckpt_fp)


    

    




INFO:tensorflow:Restoring parameters from ./train/model.ckpt-6532

In [2]:

    

# Generate disentangled latent codes

nids = 4
nobs = 8

samp_feeds = {}
samp_feeds[graph.get_tensor_by_name('samp_zi_n:0')] = nids
samp_feeds[graph.get_tensor_by_name('samp_zo_n:0')] = nobs
samp_fetches = {}
samp_fetches['zis'] = graph.get_tensor_by_name('samp_zi:0')
samp_fetches['zos'] = graph.get_tensor_by_name('samp_zo:0')

_samp_fetches = sess.run(samp_fetches, samp_feeds)

print _samp_fetches['zis'].shape
print _samp_fetches['zos'].shape


    

    




(4, 50)
(8, 50)

In [3]:

    

# Generate grid of images from latent codes

feeds = {}
feeds[graph.get_tensor_by_name('zi:0')] = _samp_fetches['zis']
feeds[graph.get_tensor_by_name('zo:0')] = _samp_fetches['zos']
fetches = {}
fetches['G_z_grid'] = graph.get_tensor_by_name('G_z_grid:0')
fetches['G_z_grid_prev'] = graph.get_tensor_by_name('G_z_grid_prev:0')

_fetches = sess.run(fetches, feeds)

print _fetches['G_z_grid'].shape
print _fetches['G_z_grid_prev'].shape


    

    




(4, 8, 64, 64, 3)
(256, 512, 3)

In [ ]:

    

# Preview image

from cStringIO import StringIO
from IPython import display
import numpy as np
import PIL.Image

def display_img(a):
    f = StringIO()
    PIL.Image.fromarray(a).save(f, 'png')
    display.display(display.Image(data=f.getvalue()))

display_img(_fetches['G_z_grid_prev'])