def estimate_image_tensorflow(msi, model_checkpoint_dir):
# estimate parameters
collapsed_msi = imgmani.collapse_image(msi.get_image())
# in case of nan values: set to 0
collapsed_msi[np.isnan(collapsed_msi)] = 0.
collapsed_msi[np.isinf(collapsed_msi)] = 0.
tf.reset_default_graph()
keep_prob = tf.placeholder("float")
nr_wavelengths = len(msi.get_wavelengths())
x = tf.placeholder("float", [None, nr_wavelengths, 1, 1])
x_test_image = np.reshape(msi.get_image(), [-1, nr_wavelengths, 1, 1])
# Construct the desired model
# pred, regularizers = multilayer_perceptron(x, nr_wavelengths, 100, 1,
# keep_prob)
pred = cnn(x, 1, keep_prob)
# Initializing the variables
init = tf.initialize_all_variables()
saver = tf.train.Saver()
with tf.Session() as sess:
sess.run(tf.initialize_all_variables())
# restore model:
ckpt = tf.train.get_checkpoint_state(model_checkpoint_dir)
if ckpt and ckpt.model_checkpoint_path:
saver.restore(sess, ckpt.model_checkpoint_path)
start = time.time()
estimated_parameters = pred.eval({x: x_test_image,
keep_prob:1.0})
end = time.time()
estimation_time = end - start
logging.info("time necessary for estimating image parameters: " +
str(estimation_time) + "s")
# restore shape
feature_dimension = 1
if len(estimated_parameters.shape) > 1:
feature_dimension = estimated_parameters.shape[-1]
estimated_paramters_as_image = np.reshape(
estimated_parameters, (msi.get_image().shape[0],
msi.get_image().shape[1],
feature_dimension))
# save as sitk nrrd.
sitk_img = sitk.GetImageFromArray(estimated_paramters_as_image,
isVector=True)
return sitk_img, estimation_time
def estimate_image_tensorflow(msi, model_checkpoint_dir):
# estimate parameters
collapsed_msi = imgmani.collapse_image(msi.get_image())
# in case of nan values: set to 0
collapsed_msi[np.isnan(collapsed_msi)] = 0.
collapsed_msi[np.isinf(collapsed_msi)] = 0.
tf.reset_default_graph()
keep_prob = tf.placeholder("float")
nr_wavelengths = len(msi.get_wavelengths())
x = tf.placeholder("float", [None, nr_wavelengths, 1, 1])
x_test_image = np.reshape(msi.get_image(), [-1, nr_wavelengths, 1, 1])
# Construct the desired model
# pred, regularizers = multilayer_perceptron(x, nr_wavelengths, 100, 1,
# keep_prob)
pred = cnn(x, 1, keep_prob)
# Initializing the variables
init = tf.initialize_all_variables()
saver = tf.train.Saver()
with tf.Session() as sess:
sess.run(tf.initialize_all_variables())
# restore model:
ckpt = tf.train.get_checkpoint_state(model_checkpoint_dir)
if ckpt and ckpt.model_checkpoint_path:
saver.restore(sess, ckpt.model_checkpoint_path)
start = time.time()
estimated_parameters = pred.eval({x: x_test_image,
keep_prob:1.0})
end = time.time()
estimation_time = end - start
logging.info("time necessary for estimating image parameters: " +
str(estimation_time) + "s")
# restore shape
feature_dimension = 1
if len(estimated_parameters.shape) > 1:
feature_dimension = estimated_parameters.shape[-1]
estimated_paramters_as_image = np.reshape(
estimated_parameters, (msi.get_image().shape[0],
msi.get_image().shape[1],
feature_dimension))
# save as sitk nrrd.
sitk_img = sitk.GetImageFromArray(estimated_paramters_as_image,
isVector=True)
return sitk_img, estimation_time
def run(self):
# extract data from the batch
tensorflow_dataset = read_data_set(self.input()[0].path)
test_dataset = read_data_set(self.input()[1].path)
# train regressor
# Construct the desired model
# Network Parameters
nr_filters = len(sc.other["RECORDED_WAVELENGTHS"])
x = tf.placeholder("float", [None, nr_filters, 1, 1])
# Construct the desired model
keep_prob = tf.placeholder("float")
# pred, regularizers = multilayer_perceptron(x, nr_filters, 100, 1,
# keep_prob)
pred = cnn(x, 1, keep_prob)
# define parameters
learning_rate = 0.0001
training_epochs = 300
batch_size = 100
display_step = 1
# Define loss and optimizer
y = tf.placeholder("float", [None, 1])
cost = tf.reduce_mean(tf.square(pred - y))
optimizer = tf.train.AdamOptimizer(learning_rate=learning_rate).minimize(cost)
# Initializing the variables
init = tf.initialize_all_variables()
saver = tf.train.Saver() # defaults to saving all variables
# Launch the graph
with tf.Session() as sess:
sess.run(init)
# Training cycle
for epoch in range(training_epochs):
avg_cost = 0.
total_batch = int(tensorflow_dataset.num_examples/batch_size)
# Loop over all batches
for i in range(total_batch):
batch_xs, batch_ys = tensorflow_dataset.next_batch(batch_size)
# Fit training using batch data
x_image = np.reshape(batch_xs, [-1, nr_filters, 1, 1])
sess.run(optimizer, feed_dict={x: x_image, y: batch_ys,
keep_prob: 0.75})
# Compute average loss
avg_cost += sess.run(cost, feed_dict={x: x_image, y: batch_ys,
keep_prob: 1.0})/total_batch
# Display logs per epoch step
if epoch % display_step == 0:
print "Epoch:", '%04d' % (epoch+1), "cost=", "{:.9f}".format(avg_cost)
# Test model
accuracy = tf.reduce_mean(tf.cast(tf.abs(pred-y), "float"))
x_test_image = np.reshape(test_dataset.images, [-1, nr_filters, 1, 1])
print "Mean testing error:", accuracy.eval({x: x_test_image,
y: test_dataset.labels,
keep_prob:1.0})
print "Optimization Finished!"
saver.save(sess, self.output().path)
def run(self):
# extract data from the batch
tensorflow_dataset = read_data_set(self.input()[0].path)
test_dataset = read_data_set(self.input()[1].path)
# train regressor
# Construct the desired model
# Network Parameters
nr_filters = len(sc.other["RECORDED_WAVELENGTHS"])
x = tf.placeholder("float", [None, nr_filters, 1, 1])
# Construct the desired model
keep_prob = tf.placeholder("float")
# pred, regularizers = multilayer_perceptron(x, nr_filters, 100, 1,
# keep_prob)
pred = cnn(x, 1, keep_prob)
# define parameters
learning_rate = 0.0001
training_epochs = 300
batch_size = 100
display_step = 1
# Define loss and optimizer
y = tf.placeholder("float", [None, 1])
cost = tf.reduce_mean(tf.square(pred - y))
optimizer = tf.train.AdamOptimizer(
learning_rate=learning_rate).minimize(cost)
# Initializing the variables
init = tf.initialize_all_variables()
saver = tf.train.Saver() # defaults to saving all variables
# Launch the graph
with tf.Session() as sess:
sess.run(init)
# Training cycle
for epoch in range(training_epochs):
avg_cost = 0.
total_batch = int(tensorflow_dataset.num_examples / batch_size)
# Loop over all batches
for i in range(total_batch):
batch_xs, batch_ys = tensorflow_dataset.next_batch(
batch_size)
# Fit training using batch data
x_image = np.reshape(batch_xs, [-1, nr_filters, 1, 1])
sess.run(optimizer,
feed_dict={
x: x_image,
y: batch_ys,
keep_prob: 0.75
})
# Compute average loss
avg_cost += sess.run(cost,
feed_dict={
x: x_image,
y: batch_ys,
keep_prob: 1.0
}) / total_batch
# Display logs per epoch step
if epoch % display_step == 0:
print "Epoch:", '%04d' % (
epoch + 1), "cost=", "{:.9f}".format(avg_cost)
# Test model
accuracy = tf.reduce_mean(tf.cast(tf.abs(pred - y), "float"))
x_test_image = np.reshape(test_dataset.images,
[-1, nr_filters, 1, 1])
print "Mean testing error:", accuracy.eval({
x: x_test_image,
y: test_dataset.labels,
keep_prob: 1.0
})
print "Optimization Finished!"
saver.save(sess, self.output().path)
