# Placeholders
X = tf.placeholder_with_default(X_def, shape=[None, 288, 288, 1])
y = tf.placeholder_with_default(y_def, shape=[None, 288, 288, 1])
# cast to float and scale input data
X_adj = tf.cast(X, dtype=tf.float32)
X_adj = _scale_input_data(X_adj,
contrast=contrast,
mu=127.0,
scale=255.0)
# optional online data augmentation
if distort:
X_adj, y = augment(X_adj,
y,
horizontal_flip=True,
augment_labels=True,
vertical_flip=True,
mixup=0)
# Convolutional layer 1
with tf.name_scope('conv1') as scope:
conv1 = tf.layers.conv2d(
X_adj, # Input data
filters=32, # 32 filters
kernel_size=(3, 3), # Kernel size: 5x5
strides=(2, 2), # Stride: 2
padding='SAME', # "same" padding
activation=None, # None
kernel_initializer=tf.truncated_normal_initializer(stddev=5e-2,
seed=100),
kernel_regularizer=tf.contrib.layers.l2_regularizer(scale=lamC),
with tf.device('/cpu:0'):
image, label = read_and_decode_single_example(train_files, label_type=how, normalize=False, distort=False)
X_def, y_def = tf.train.shuffle_batch([image, label], batch_size=batch_size, capacity=2000,
seed=None,
min_after_dequeue=1000)
# Placeholders
X = tf.placeholder_with_default(X_def, shape=[None, 288, 288, 1])
y = tf.placeholder_with_default(y_def, shape=[None, 288, 288, 1])
X_fl = tf.cast(X, tf.float32)
# optional online data augmentation
if distort:
X_dis, y_adj = augment(X_fl, y, horizontal_flip=True, augment_labels=True, vertical_flip=True, mixup=0)
else:
y_adj = y
X_dis = X_fl
# cast to float and scale input data
X_adj = _scale_input_data(X_dis, contrast=contrast, mu=127.0, scale=255.0)
# Convolutional layer 1
with tf.name_scope('conv1') as scope:
conv1 = tf.layers.conv2d(
X_adj, # Input data
filters=32, # 32 filters
kernel_size=(3, 3), # Kernel size: 5x5
strides=(2, 2), # Stride: 2
padding='SAME', # "same" padding
