def sample(self, nb_samples):
sampled_character_folders = random.sample(self.character_folders,
nb_samples)
random.shuffle(sampled_character_folders)
example_inputs = np.zeros(
(self.batch_size, nb_samples * self.nb_samples_per_class,
np.prod(self.img_size)),
dtype=np.float32)
example_outputs = np.zeros(
(self.batch_size, nb_samples * self.nb_samples_per_class),
dtype=np.float32
) #notice hardcoded np.float32 here and above, change it to something else in tf
for i in range(self.batch_size):
labels_and_images = get_shuffled_images(
sampled_character_folders,
range(nb_samples),
nb_samples=self.nb_samples_per_class)
sequence_length = len(labels_and_images)
labels, image_files = zip(*labels_and_images)
angles = np.random.uniform(-self.max_rotation,
self.max_rotation,
size=sequence_length)
shifts = np.random.uniform(-self.max_shift,
self.max_shift,
size=sequence_length)
example_inputs[i] = np.asarray([load_transform(filename, angle=angle, s=shift, size=self.img_size).flatten() \
for (filename, angle, shift) in zip(image_files, angles, shifts)], dtype=np.float32)
example_outputs[i] = np.asarray(labels, dtype=np.int32)
return example_inputs, example_outputs
def sample(self, nb_samples):
sampled_character_folders = random.sample(self.character_folders, nb_samples)
random.shuffle(sampled_character_folders)
example_inputs = np.zeros((self.batch_size, nb_samples * self.nb_samples_per_class, np.prod(self.img_size)), dtype=np.float32)
example_outputs = np.zeros((self.batch_size, nb_samples * self.nb_samples_per_class), dtype=np.float32) #notice hardcoded np.float32 here and above, change it to something else in tf
for i in range(self.batch_size):
labels_and_images = get_shuffled_images(sampled_character_folders, range(nb_samples), nb_samples=self.nb_samples_per_class)
sequence_length = len(labels_and_images)
labels, image_files = zip(*labels_and_images)
angles = np.random.uniform(-self.max_rotation, self.max_rotation, size=sequence_length)
shifts = np.random.uniform(-self.max_shift, self.max_shift, size=sequence_length)
example_inputs[i] = np.asarray([load_transform(filename, angle=angle, s=shift, size=self.img_size).flatten() \
for (filename, angle, shift) in zip(image_files, angles, shifts)], dtype=np.float32)
example_outputs[i] = np.asarray(labels, dtype=np.int32)
return example_inputs, example_outputs
def startom():
noise_input2 = tf.placeholder(tf.float32, shape=[None, latent_dim2])
decoder2 = tf.matmul(tf.concat([noise_input2,y2], 1), weights2['decoder_h1']) + biases2['decoder_b1']
decoder2 = tf.nn.tanh(decoder2)
decoder2 = tf.matmul(decoder2, weights2['decoder_out']) + biases2['decoder_out']
decoder2 = tf.nn.sigmoid(decoder2)
n = 9
noise_dim=2
filename='/var/www/html/search/images/input.png'
example_inputs = np.asarray([load_transform(filename, size=(28,28)).flatten() \
], dtype=np.float32)
canvas_recon0 = np.empty((28 * (1), 28 * 1))
canvas_recon1 = np.empty((28 * (1), 28 * 1))
canvas_recon2 = np.empty((28 * (1), 28 * 1))
canvas_recon3 = np.empty((28 * (1), 28 * 1))
canvas_recon4 = np.empty((28 * (1), 28 * 1))
canvas_recon5 = np.empty((28 * (1), 28 * 1))
canvas_recon6 = np.empty((28 * (1), 28 * 1))
canvas_recon7 = np.empty((28 * (1), 28 * 1))
canvas_recon8 = np.empty((28 * (1), 28 * 1))
for i in range(n):
ztest = np.random.uniform(-1., 1., size=[1, noise_dim])
g = sess.run(decoder2, feed_dict={noise_input2: ztest,y2:example_inputs })
g = (g + 1.) / 2.
g = -1 * (g - 1)
if i==0:
canvas_recon0[ 0:(1) * 28, 0:(1) * 28] = g.reshape([28, 28])
plt.figure(figsize=(1, 1))
plt.axis('off')
plt.imshow(canvas_recon0, origin="upper", cmap="gray")
plt.savefig('/var/www/html/search/images/out%d.png'%(i+1) )
if i==1:
canvas_recon1[ 0:(1) * 28, 0:(1) * 28] = g.reshape([28, 28])
plt.figure(figsize=(1, 1))
plt.axis('off')
plt.imshow(canvas_recon1, origin="upper", cmap="gray")
plt.savefig('/var/www/html/search/images/out%d.png'%(i+1) )
if i==2:
canvas_recon2[ 0:(1) * 28, 0:(1) * 28] = g.reshape([28, 28])
plt.figure(figsize=(1, 1))
plt.axis('off')
plt.imshow(canvas_recon2, origin="upper", cmap="gray")
plt.savefig('/var/www/html/search/images/out%d.png'%(i+1) )
if i==3:
canvas_recon3[ 0:(1) * 28, 0:(1) * 28] = g.reshape([28, 28])
plt.figure(figsize=(1, 1))
plt.axis('off')
plt.imshow(canvas_recon3, origin="upper", cmap="gray")
plt.savefig('/var/www/html/search/images/out%d.png'%(i+1) )
if i==4:
canvas_recon4[ 0:(1) * 28, 0:(1) * 28] = g.reshape([28, 28])
plt.figure(figsize=(1, 1))
plt.axis('off')
plt.imshow(canvas_recon4, origin="upper", cmap="gray")
plt.savefig('/var/www/html/search/images/out%d.png'%(i+1) )
if i==5:
canvas_recon5[ 0:(1) * 28, 0:(1) * 28] = g.reshape([28, 28])
plt.figure(figsize=(1, 1))
plt.axis('off')
plt.imshow(canvas_recon5, origin="upper", cmap="gray")
plt.savefig('/var/www/html/search/images/out%d.png'%(i+1) )
if i==6:
canvas_recon6[ 0:(1) * 28, 0:(1) * 28] = g.reshape([28, 28])
plt.figure(figsize=(1, 1))
plt.axis('off')
plt.imshow(canvas_recon6, origin="upper", cmap="gray")
plt.savefig('/var/www/html/search/images/out%d.png'%(i+1) )
if i==7:
canvas_recon7[ 0:(1) * 28, 0:(1) * 28] = g.reshape([28, 28])
plt.figure(figsize=(1, 1))
plt.axis('off')
plt.imshow(canvas_recon7, origin="upper", cmap="gray")
plt.savefig('/var/www/html/search/images/out%d.png'%(i+1) )
if i==8:
canvas_recon8[ 0:(1) * 28, 0:(1) * 28] = g.reshape([28, 28])
plt.figure(figsize=(1, 1))
plt.axis('off')
plt.imshow(canvas_recon8, origin="upper", cmap="gray")
plt.savefig('/var/www/html/search/images/out%d.png'%(i+1) )