train_path=train_path,
image_folder='liver',
mask_folder='mask',
target_size=img_size)
# check if pretrained weights are defined
if is_file(file_name=model_weights_name):
pretrained_weights = model_weights_name
else:
pretrained_weights = None
# build model
unett = UNet(input_size=(img_width, img_height, 1),
n_filters=64,
pretrained_weights=pretrained_weights)
unett.build()
# creating a callback, hence best weights configurations will be saved
model_checkpoint = unett.checkpoint(model_name)
# model training
# steps per epoch should be equal to number of samples in database divided by batch size
# in this case, it is 560 / 2 = 280
unett.fit_generator(train_gen,
steps_per_epoch=281,
epochs=10,
callbacks=[model_checkpoint])
#history=unett.fit(
# train_gen,
# steps_per_epoch=280,
# epochs=5,
test_path = '/home/new-ece/szc0173/liver/test/'
save_path = '/home/new-ece/szc0173/liver/result2/'
model_weights_name = 'unet_weight_model.hdf5'
# TODO: move to config .json files
img_height = 512
img_width = 512
img_size = (img_height, img_width)
model = UNet(input_size=(img_width, img_height, 1),
n_filters=64,
pretrained_weights=model_weights_name)
train_gen = train_generator(batch_size=2,
train_path=train_path,
image_folder='liver',
mask_folder='mask',
target_size=img_size)
model.build()
model_name2 = '/home/new-ece/szc0173/liver/unet_model2.hdf5'
model_checkpoint2 = model.checkpoint(model_name2)
model.fit_generator(train_gen,
steps_per_epoch=280,
epochs=10,
callbacks=[model_checkpoint2])
#history1=model.fit(
# train_gen,
# steps_per_epoch =280 ,
# epochs = 10,
# callbacks = [model_checkpoint2]
# )
# saving model weights
model.save_model('/home/new-ece/szc0173/liver/unet_weight_model2.hdf5')
def next_train_batch(batch_s, batch_count, is_first_iter):
if (batch_count == 0):
shuffle()
count = batch_s * batch_count
return images[count:(count + batch_s)], labels[count:(count + batch_s)]
tf.reset_default_graph()
X = tf.placeholder(tf.float32, [None, IMG_SIZE, IMG_SIZE, IMG_CHANNELS])
Y_ = tf.placeholder(tf.float32, [None, IMG_SIZE, IMG_SIZE, 1])
lr = tf.placeholder(tf.float32)
#
model = UNet()
model.build(X)
cross_entropy = tf.losses.sigmoid_cross_entropy(Y_, model.output)
train_step = tf.train.AdamOptimizer(lr).minimize(cross_entropy)
# init
init = tf.global_variables_initializer()
sess = tf.Session()
sess.run(init)
batch_count = 0
display_count = 1
epoch_loss = 0
best_loss = np.inf
saver = tf.train.Saver()
writer = tf.summary.FileWriter('../../logs')
writer.add_graph(tf.get_default_graph())