def test_input_pipeline(self):
Xs, Ys = dsu.tiny_imagenet_load()
n_batches = 0
batch_size = 10
with tf.Graph().as_default(), tf.Session() as sess:
batch_generator = dsu.create_input_pipeline(
Xs[:100],
batch_size=batch_size,
n_epochs=1,
shape=(64, 64, 3),
crop_shape=(64, 64, 3))
init_op = tf.group(tf.global_variables_initializer(),
tf.local_variables_initializer())
sess.run(init_op)
coord = tf.train.Coordinator()
tf.get_default_graph().finalize()
threads = tf.train.start_queue_runners(sess=sess, coord=coord)
try:
while not coord.should_stop():
batch = sess.run(batch_generator)
assert (batch.shape == (batch_size, 64, 64, 3))
n_batches += 1
except tf.errors.OutOfRangeError:
pass
finally:
coord.request_stop()
coord.join(threads)
assert (n_batches == 10)
def test_draw_train_input_pipeline(self):
Xs, ys = dsu.tiny_imagenet_load()
draw.train_input_pipeline(Xs[:100],
batch_size=20,
n_epochs=1,
A=64,
B=64,
C=3,
input_shape=(64, 64, 3))
def generate():
"""Summary
"""
# Parameters for generation
ckpt_path = 'pixelcnn'
B = None
H = 64
W = 64
C = 3
with tf.Graph().as_default(), tf.Session() as sess:
# Not actually conditioning on anything here just using the gated cnn model
net = build_conditional_pixel_cnn_model(B=B, H=H, W=W, C=C)
# Load a list of files for tiny imagenet, downloading if necessary
imagenet_files = dsu.tiny_imagenet_load()
saver = tf.train.Saver()
init_op = tf.group(tf.global_variables_initializer(),
tf.local_variables_initializer())
sess.run(init_op)
saver.restore(sess, tf.train.latest_checkpoint(ckpt_path))
import matplotlib.pyplot as plt
img = plt.imread(imagenet_files[0][1000])
from scipy.misc import imresize
og_img = imresize(img, (H, W))
img = og_img.copy()
# Zero out bottom half of image and let's try to synthesize it
img[H // 2:, :, :] = 0
for h_i in range(H // 2, H):
for w_i in range(W):
for c_i in range(C):
print(h_i, w_i, c_i, end='\r')
X = img.copy()
preds = sess.run(
net['sampled_preds'],
feed_dict={net['X']: X[np.newaxis]})
X = preds.reshape((1, H, W, C)).astype(np.uint8)
img[h_i, w_i, c_i] = X[0, h_i, w_i, c_i]
fig, axs = plt.subplots(1, 2)
axs[0].imshow(og_img)
axs[1].imshow(img)
def generate():
"""Summary
"""
# Parameters for generation
ckpt_path = 'pixelcnn'
B = None
H = 64
W = 64
C = 3
with tf.Graph().as_default(), tf.Session() as sess:
# Not actually conditioning on anything here just using the gated cnn model
net = build_conditional_pixel_cnn_model(B=B, H=H, W=W, C=C)
# Load a list of files for tiny imagenet, downloading if necessary
imagenet_files = dsu.tiny_imagenet_load()
saver = tf.train.Saver()
init_op = tf.group(tf.global_variables_initializer(),
tf.local_variables_initializer())
sess.run(init_op)
saver.restore(sess, tf.train.latest_checkpoint(ckpt_path))
import matplotlib.pyplot as plt
img = plt.imread(imagenet_files[0][1000])
from scipy.misc import imresize
og_img = imresize(img, (H, W))
img = og_img.copy()
# Zero out bottom half of image and let's try to synthesize it
img[H // 2:, :, :] = 0
for h_i in range(H // 2, H):
for w_i in range(W):
for c_i in range(C):
print(h_i, w_i, c_i, end='\r')
X = img.copy()
preds = sess.run(net['sampled_preds'],
feed_dict={net['X']: X[np.newaxis]})
X = preds.reshape((1, H, W, C)).astype(np.uint8)
img[h_i, w_i, c_i] = X[0, h_i, w_i, c_i]
fig, axs = plt.subplots(1, 2)
axs[0].imshow(og_img)
axs[1].imshow(img)
def TINYIMAGENET(path='./tiny_imagenet/'):
"""Attempt to load the files of the Tiny ImageNet dataset.
http://cs231n.stanford.edu/tiny-imagenet-200.zip
https://tiny-imagenet.herokuapp.com/
Parameters
----------
path : str, optional
Directory where the dataset can be found or else will be placed.
Returns
-------
files : list
List of file paths to the dataset.
labels : list
List of labels for each file (only training files have labels)
"""
files, labels = tiny_imagenet_load(path)
return files, labels
def TINYIMAGENET(path='./tiny_imagenet/'):
"""Attempt to load the files of the Tiny ImageNet dataset.
http://cs231n.stanford.edu/tiny-imagenet-200.zip
https://tiny-imagenet.herokuapp.com/
Parameters
----------
path : str, optional
Directory where the dataset can be found or else will be placed.
Returns
-------
files : list
List of file paths to the dataset.
labels : list
List of labels for each file (only training files have labels)
"""
files, labels = tiny_imagenet_load(path)
return files, labels
def train_tiny_imagenet(ckpt_path='pixelcnn',
n_epochs=1000,
save_step=100,
write_step=25,
B=32,
H=64,
W=64,
C=3):
"""Summary
Parameters
----------
ckpt_path : str, optional
Description
n_epochs : int, optional
Description
save_step : int, optional
Description
write_step : int, optional
Description
B : int, optional
Description
H : int, optional
Description
W : int, optional
Description
C : int, optional
Description
"""
ckpt_name = os.path.join(ckpt_path, 'pixelcnn.ckpt')
with tf.Graph().as_default(), tf.Session() as sess:
# Not actually conditioning on anything here just using the gated cnn model
net = build_conditional_pixel_cnn_model(B=B, H=H, W=W, C=C)
# build the optimizer (this will take a while!)
optimizer = tf.train.AdamOptimizer(
learning_rate=0.001).minimize(net['cost'])
# Load a list of files for tiny imagenet, downloading if necessary
imagenet_files = dsu.tiny_imagenet_load()
# Create a threaded image pipeline which will load/shuffle/crop/resize
batch = dsu.create_input_pipeline(
imagenet_files[0],
batch_size=B,
n_epochs=n_epochs,
shape=[64, 64, 3],
crop_shape=[H, W, C],
crop_factor=1.0,
n_threads=8)
saver = tf.train.Saver()
writer = tf.summary.FileWriter(ckpt_path)
init_op = tf.group(tf.global_variables_initializer(),
tf.local_variables_initializer())
sess.run(init_op)
# This will handle our threaded image pipeline
coord = tf.train.Coordinator()
# Ensure no more changes to graph
tf.get_default_graph().finalize()
# Start up the queues for handling the image pipeline
threads = tf.train.start_queue_runners(sess=sess, coord=coord)
if os.path.exists(ckpt_name + '.index') or os.path.exists(ckpt_name):
saver.restore(sess, ckpt_name)
saver.restore(sess, tf.train.latest_checkpoint(ckpt_path))
epoch_i = 0
batch_i = 0
try:
while not coord.should_stop() and epoch_i < n_epochs:
batch_i += 1
batch_xs = sess.run(batch)
train_cost = sess.run(
[net['cost'], optimizer], feed_dict={net['X']: batch_xs})[0]
print(batch_i, train_cost)
if batch_i % write_step == 0:
summary = sess.run(
net['summaries'], feed_dict={net['X']: batch_xs})
writer.add_summary(summary, batch_i)
if batch_i % save_step == 0:
# Save the variables to disk. Don't write the meta graph
# since we can use the code to create it, and it takes a long
# time to create the graph since it is so deep
saver.save(
sess,
ckpt_name,
global_step=batch_i,
write_meta_graph=True)
except tf.errors.OutOfRangeError:
print('Done.')
finally:
# One of the threads has issued an exception. So let's tell all the
# threads to shutdown.
coord.request_stop()
# Wait until all threads have finished.
coord.join(threads)
def test_tiny_imagenet_loads(self):
Xs, Ys = dsu.tiny_imagenet_load()
assert (len(Xs) == 120000)
assert (Xs[0].endswith('n02226429_40.JPEG'))
assert (Ys[0] == 'grasshopper, hopper')
def train_tiny_imagenet():
"""Summary
"""
net = build_pixel_rnn_basic_model()
# build the optimizer (this will take a while!)
optimizer = tf.train.AdamOptimizer(
learning_rate=0.001).minimize(net['cost'])
# Load a list of files for tiny imagenet, downloading if necessary
imagenet_files = dsu.tiny_imagenet_load()
# Create a threaded image pipeline which will load/shuffle/crop/resize
batch = dsu.create_input_pipeline(
imagenet_files,
batch_size=B,
n_epochs=n_epochs,
shape=[64, 64, 3],
crop_shape=[32, 32, 3],
crop_factor=0.5,
n_threads=8)
sess = tf.Session()
saver = tf.train.Saver()
init_op = tf.group(tf.global_variables_initializer(),
tf.local_variables_initializer())
sess.run(init_op)
# This will handle our threaded image pipeline
coord = tf.train.Coordinator()
# Ensure no more changes to graph
tf.get_default_graph().finalize()
# Start up the queues for handling the image pipeline
threads = tf.train.start_queue_runners(sess=sess, coord=coord)
if os.path.exists(ckpt_name + '.index') or os.path.exists(ckpt_name):
saver.restore(sess, ckpt_name)
saver.restore(sess, tf.train.latest_checkpoint('./'))
epoch_i = 0
batch_i = 0
save_step = 100
try:
while not coord.should_stop() and epoch_i < n_epochs:
batch_i += 1
batch_xs = sess.run(batch)
train_cost = sess.run(
[net['cost'], optimizer], feed_dict={net['X']: batch_xs})[0]
print(batch_i, train_cost)
if batch_i % save_step == 0:
# Save the variables to disk. Don't write the meta graph
# since we can use the code to create it, and it takes a long
# time to create the graph since it is so deep
saver.save(
sess,
ckpt_name,
global_step=batch_i,
write_meta_graph=False)
except tf.errors.OutOfRangeError:
print('Done.')
finally:
# One of the threads has issued an exception. So let's tell all the
# threads to shutdown.
coord.request_stop()
# Wait until all threads have finished.
coord.join(threads)
# Clean up the session.
sess.close()
def test_draw_train_input_pipeline(self):
Xs, ys = dsu.tiny_imagenet_load()
draw.train_input_pipeline(
Xs[:100], batch_size=20, n_epochs=1, A=64, B=64, C=3,
input_shape=(64, 64, 3))
def train_tiny_imagenet(ckpt_path='pixelcnn',
n_epochs=1000,
save_step=100,
write_step=25,
B=32,
H=64,
W=64,
C=3):
"""Summary
Parameters
----------
ckpt_path : str, optional
Description
n_epochs : int, optional
Description
save_step : int, optional
Description
write_step : int, optional
Description
B : int, optional
Description
H : int, optional
Description
W : int, optional
Description
C : int, optional
Description
"""
ckpt_name = os.path.join(ckpt_path, 'pixelcnn.ckpt')
with tf.Graph().as_default(), tf.Session() as sess:
# Not actually conditioning on anything here just using the gated cnn model
net = build_conditional_pixel_cnn_model(B=B, H=H, W=W, C=C)
# build the optimizer (this will take a while!)
optimizer = tf.train.AdamOptimizer(learning_rate=0.001).minimize(
net['cost'])
# Load a list of files for tiny imagenet, downloading if necessary
imagenet_files = dsu.tiny_imagenet_load()
# Create a threaded image pipeline which will load/shuffle/crop/resize
batch = dsu.create_input_pipeline(imagenet_files[0],
batch_size=B,
n_epochs=n_epochs,
shape=[64, 64, 3],
crop_shape=[H, W, C],
crop_factor=1.0,
n_threads=8)
saver = tf.train.Saver()
writer = tf.summary.FileWriter(ckpt_path)
init_op = tf.group(tf.global_variables_initializer(),
tf.local_variables_initializer())
sess.run(init_op)
# This will handle our threaded image pipeline
coord = tf.train.Coordinator()
# Ensure no more changes to graph
tf.get_default_graph().finalize()
# Start up the queues for handling the image pipeline
threads = tf.train.start_queue_runners(sess=sess, coord=coord)
if os.path.exists(ckpt_name + '.index') or os.path.exists(ckpt_name):
saver.restore(sess, ckpt_name)
saver.restore(sess, tf.train.latest_checkpoint(ckpt_path))
epoch_i = 0
batch_i = 0
try:
while not coord.should_stop() and epoch_i < n_epochs:
batch_i += 1
batch_xs = sess.run(batch)
train_cost = sess.run([net['cost'], optimizer],
feed_dict={net['X']: batch_xs})[0]
print(batch_i, train_cost)
if batch_i % write_step == 0:
summary = sess.run(net['summaries'],
feed_dict={net['X']: batch_xs})
writer.add_summary(summary, batch_i)
if batch_i % save_step == 0:
# Save the variables to disk. Don't write the meta graph
# since we can use the code to create it, and it takes a long
# time to create the graph since it is so deep
saver.save(sess,
ckpt_name,
global_step=batch_i,
write_meta_graph=True)
except tf.errors.OutOfRangeError:
print('Done.')
finally:
# One of the threads has issued an exception. So let's tell all the
# threads to shutdown.
coord.request_stop()
# Wait until all threads have finished.
coord.join(threads)