In [1]:
%matplotlib inline
import numpy as np
import tensorflow as tf
import matplotlib.pyplot as plt
from tensorflow.examples.tutorials.mnist import input_data
mnist = input_data.read_data_sets('MNIST_data', validation_size=0)
Extracting MNIST_data/train-images-idx3-ubyte.gz Extracting MNIST_data/train-labels-idx1-ubyte.gz Extracting MNIST_data/t10k-images-idx3-ubyte.gz Extracting MNIST_data/t10k-labels-idx1-ubyte.gz
In [2]:
img = mnist.train.images[2]
plt.imshow(img.reshape((28, 28)), cmap='Greys_r')
Out[2]:
<matplotlib.image.AxesImage at 0x7f2c0a4dc5f8>
In [0]:
inputs_ = tf.placeholder(tf.float32, (None, 28, 28, 1), name='inputs')
targets_ = tf.placeholder(tf.float32, (None, 28, 28, 1), name='targets')
### Encoder
conv1 = tf.layers.conv2d(inputs_, 16, (3,3), padding='same', activation=tf.nn.relu)
# Now 28x28x16
maxpool1 = tf.layers.max_pooling2d(conv1, (2,2), (2,2), padding='same')
# Now 14x14x16
conv2 = tf.layers.conv2d(maxpool1, 8, (3,3), padding='same', activation=tf.nn.relu)
# Now 14x14x8
maxpool2 = tf.layers.max_pooling2d(conv2, (2,2), (2,2), padding='same')
# Now 7x7x8
conv3 = tf.layers.conv2d(maxpool2, 8, (3,3), padding='same', activation=tf.nn.relu)
# Now 7x7x8
encoded = tf.layers.max_pooling2d(conv3, (2,2), (2,2), padding='same')
# Now 4x4x8
### Decoder
upsample1 = tf.image.resize_nearest_neighbor(encoded, (7,7))
# Now 7x7x8
conv4 = tf.layers.conv2d(upsample1, 8, (3,3), padding='same', activation=tf.nn.relu)
# Now 7x7x8
upsample2 = tf.image.resize_nearest_neighbor(conv4, (14,14))
# Now 14x14x8
conv5 = tf.layers.conv2d(upsample2, 8, (3,3), padding='same', activation=tf.nn.relu)
# Now 14x14x8
upsample3 = tf.image.resize_nearest_neighbor(conv5, (28,28))
# Now 28x28x8
conv6 = tf.layers.conv2d(upsample3, 16, (3,3), padding='same', activation=tf.nn.relu)
# Now 28x28x16
logits = tf.layers.conv2d(conv6, 1, (3,3), padding='same', activation=None)
#Now 28x28x1
decoded = tf.nn.sigmoid(logits, name='decoded')
loss = tf.nn.sigmoid_cross_entropy_with_logits(labels=targets_, logits=logits)
cost = tf.reduce_mean(loss)
opt = tf.train.AdamOptimizer(0.001).minimize(cost)
In [0]:
sess = tf.Session()
In [5]:
epochs = 20
batch_size = 200
sess.run(tf.global_variables_initializer())
for e in range(epochs):
loss=[]
for ii in range(mnist.train.num_examples//batch_size):
batch = mnist.train.next_batch(batch_size)
imgs = batch[0].reshape((-1, 28, 28, 1))
batch_cost, _ = sess.run([cost, opt], feed_dict={inputs_: imgs,
targets_: imgs})
loss.append(batch_cost)
training_loss=np.mean(batch_cost)
print("Epoch: {}/{}...".format(e+1, epochs),
"Training loss: {:.4f}".format(training_loss))
Epoch: 1/20... Training loss: 0.1633 Epoch: 2/20... Training loss: 0.1351 Epoch: 3/20... Training loss: 0.1288 Epoch: 4/20... Training loss: 0.1199 Epoch: 5/20... Training loss: 0.1118 Epoch: 6/20... Training loss: 0.1169 Epoch: 7/20... Training loss: 0.1090 Epoch: 8/20... Training loss: 0.1083 Epoch: 9/20... Training loss: 0.1090 Epoch: 10/20... Training loss: 0.1067 Epoch: 11/20... Training loss: 0.1082 Epoch: 12/20... Training loss: 0.1055 Epoch: 13/20... Training loss: 0.1006 Epoch: 14/20... Training loss: 0.1043 Epoch: 15/20... Training loss: 0.1003 Epoch: 16/20... Training loss: 0.1007 Epoch: 17/20... Training loss: 0.1002 Epoch: 18/20... Training loss: 0.1008 Epoch: 19/20... Training loss: 0.0995 Epoch: 20/20... Training loss: 0.1032
In [6]:
fig, axes = plt.subplots(nrows=2, ncols=10, sharex=True, sharey=True, figsize=(20,4))
in_imgs = mnist.test.images[:10]
reconstructed = sess.run(decoded, feed_dict={inputs_: in_imgs.reshape((10, 28, 28, 1))})
for images, row in zip([in_imgs, reconstructed], axes):
for img, ax in zip(images, row):
ax.imshow(img.reshape((28, 28)), cmap='Greys_r')
ax.get_xaxis().set_visible(False)
ax.get_yaxis().set_visible(False)
fig.tight_layout(pad=0.1)
In [0]:
sess.close()
In [0]:
# Denoising
In [0]:
inputs_ = tf.placeholder(tf.float32, (None, 28, 28, 1), name='inputs')
targets_ = tf.placeholder(tf.float32, (None, 28, 28, 1), name='targets')
### Encoder
conv1 = tf.layers.conv2d(inputs_, 32, (3,3), padding='same', activation=tf.nn.relu)
# Now 28x28x32
maxpool1 = tf.layers.max_pooling2d(conv1, (2,2), (2,2), padding='same')
# Now 14x14x32
conv2 = tf.layers.conv2d(maxpool1, 32, (3,3), padding='same', activation=tf.nn.relu)
# Now 14x14x32
maxpool2 = tf.layers.max_pooling2d(conv2, (2,2), (2,2), padding='same')
# Now 7x7x32
conv3 = tf.layers.conv2d(maxpool2, 16, (3,3), padding='same', activation=tf.nn.relu)
# Now 7x7x16
encoded = tf.layers.max_pooling2d(conv3, (2,2), (2,2), padding='same')
# Now 4x4x16
### Decoder
upsample1 = tf.image.resize_nearest_neighbor(encoded, (7,7))
# Now 7x7x16
conv4 = tf.layers.conv2d(upsample1, 16, (3,3), padding='same', activation=tf.nn.relu)
# Now 7x7x16
upsample2 = tf.image.resize_nearest_neighbor(conv4, (14,14))
# Now 14x14x16
conv5 = tf.layers.conv2d(upsample2, 32, (3,3), padding='same', activation=tf.nn.relu)
# Now 14x14x32
upsample3 = tf.image.resize_nearest_neighbor(conv5, (28,28))
# Now 28x28x32
conv6 = tf.layers.conv2d(upsample3, 32, (3,3), padding='same', activation=tf.nn.relu)
# Now 28x28x32
logits = tf.layers.conv2d(conv6, 1, (3,3), padding='same', activation=None)
#Now 28x28x1
decoded = tf.nn.sigmoid(logits, name='decoded')
loss = tf.nn.sigmoid_cross_entropy_with_logits(labels=targets_, logits=logits)
cost = tf.reduce_mean(loss)
opt = tf.train.AdamOptimizer(0.001).minimize(cost)
In [10]:
sess = tf.Session()
epochs = 20
batch_size = 200
# Set's how much noise we're adding to the MNIST images
noise_factor = 0.5
sess.run(tf.global_variables_initializer())
for e in range(epochs):
loss=[]
for ii in range(mnist.train.num_examples//batch_size):
batch = mnist.train.next_batch(batch_size)
# Get images from the batch
imgs = batch[0].reshape((-1, 28, 28, 1))
# Add random noise to the input images
noisy_imgs = imgs + noise_factor * np.random.randn(*imgs.shape)
# Clip the images to be between 0 and 1
noisy_imgs = np.clip(noisy_imgs, 0., 1.)
# Noisy images as inputs, original images as targets
batch_cost, _ = sess.run([cost, opt], feed_dict={inputs_: noisy_imgs,
targets_: imgs})
loss.append(batch_cost)
training_loss=np.mean(batch_cost)
print("Epoch: {}/{}...".format(e+1, epochs),
"Training loss: {:.4f}".format(training_loss))
Epoch: 1/20... Training loss: 0.1694 Epoch: 2/20... Training loss: 0.1427 Epoch: 3/20... Training loss: 0.1280 Epoch: 4/20... Training loss: 0.1264 Epoch: 5/20... Training loss: 0.1245 Epoch: 6/20... Training loss: 0.1204 Epoch: 7/20... Training loss: 0.1202 Epoch: 8/20... Training loss: 0.1119 Epoch: 9/20... Training loss: 0.1143 Epoch: 10/20... Training loss: 0.1159 Epoch: 11/20... Training loss: 0.1138 Epoch: 12/20... Training loss: 0.1147 Epoch: 13/20... Training loss: 0.1107 Epoch: 14/20... Training loss: 0.1082 Epoch: 15/20... Training loss: 0.1093 Epoch: 16/20... Training loss: 0.1085 Epoch: 17/20... Training loss: 0.1133 Epoch: 18/20... Training loss: 0.1092 Epoch: 19/20... Training loss: 0.1051 Epoch: 20/20... Training loss: 0.1059
In [11]:
# Check out performance
fig, axes = plt.subplots(nrows=2, ncols=10, sharex=True, sharey=True, figsize=(20,4))
in_imgs = mnist.test.images[:10]
noisy_imgs = in_imgs + noise_factor * np.random.randn(*in_imgs.shape)
noisy_imgs = np.clip(noisy_imgs, 0., 1.)
reconstructed = sess.run(decoded, feed_dict={inputs_: noisy_imgs.reshape((10, 28, 28, 1))})
for images, row in zip([noisy_imgs, reconstructed], axes):
for img, ax in zip(images, row):
ax.imshow(img.reshape((28, 28)), cmap='Greys_r')
ax.get_xaxis().set_visible(False)
ax.get_yaxis().set_visible(False)
fig.tight_layout(pad=0.1)
