def main():
# # no augmentation
# from losses import dice
# loss = dice.dice_loss
# train, val = data_g.get_train_val_iterators(aug=None)
# model = cnn.get_model(None, None, 3, do_compile=False)
# model.compile(optimizer='adam', loss=loss,
# metrics=['accuracy', f1, tf.keras.metrics.MeanIoU(num_classes=2)])
# model_name = 'cnn_dice_EXTDATA_augmentation_none'
# train_sub(model, model_name, train, val, epochs=100, verbose=2)
# small augmentation
from losses import dice
loss = dice.dice_loss
train, val = data_g.get_train_val_iterators(aug='small')
#get model
model = cnn.get_model(None, None, 3, do_compile=False)
model.compile(
optimizer='adam',
loss=loss,
metrics=['accuracy', f1,
tf.keras.metrics.MeanIoU(num_classes=2)])
model_name = 'cnn_dice_EXTDATA_augmentation_small'
#train model
train_sub(model, model_name, train, val, epochs=100, verbose=2)
def get_model(base_model,
dataset_name=False,
compile=True,
weights=None,
epsilon=1e-8,
teacher_epsilon=1e-3,
init_temp=2.5):
"""Take an uncompiled model and return model compiled for ENDD.
Warning: This function works in place. Model is returned only for
conveniance.
"""
if isinstance(base_model, str):
if not dataset_name:
raise ValueError(
'dataset_name must be provided if base_model is given by name.'
)
if base_model == 'cnn':
base_model = cnn.get_model(dataset_name,
compile=False,
softmax=False)
elif base_model == 'vgg':
base_model = vgg.get_model(dataset_name,
compile=False,
softmax=False)
else:
raise ValueError(
"""Base model {} not recognized, make sure it has been added
to endd.py, or pass a Keras model object as base model instead."""
)
if weights:
saveload.load_weights(base_model, weights)
if compile:
base_model.compile(optimizer='adam',
loss=losses.DirichletEnDDLoss(
init_temp=init_temp,
epsilon=epsilon,
ensemble_epsilon=teacher_epsilon))
return base_model
def main():
x1, y1 = data.get_training_data()
x2, y2 = data.get_training_data2()
x = np.concatenate((x1, x2), axis=0)
y = np.concatenate((y1, y2), axis=0)
#x, y = data.augment_data(x, y)
# x, y = data.augment_data(x,y)
# # crossentropy
# model = unet.get_model(None, None, 3, do_compile=False)
# model.compile(optimizer='adam', loss='binary_crossentropy',
# metrics=['accuracy', tf.keras.metrics.MeanIoU(num_classes=2)])
# model_name = 'u_net_cross_entropy_test'
# train_sub(model, model_name, x, y, (x_test, y_test), epochs=100)
# # focal
# from losses import focal
# loss = focal.focal_loss
# model_name = 'u_net_focal_loss'
# model = unet.get_model(None, None, 3, do_compile=False)
# model.compile(optimizer='adam', loss=loss,
# metrics=['accuracy', tf.keras.metrics.MeanIoU(num_classes=2)])
# train_sub(model, model_name, x, y, epochs=1)
# dice
from losses import dice
loss = dice.dice_loss
model_name = 'cnn_dice_EXTDATA_100e_nostop'
model = cnn.get_model(None, None, 3, do_compile=False)
model.compile(
optimizer='adam',
loss=loss,
metrics=['accuracy',
tf.keras.metrics.MeanIoU(num_classes=2), f1])
train_sub(model, model_name, x, y, epochs=100)
import settings
from models import cnn
from utils import saveload
from utils import datasets
# Need these settings for GPU to work on my computer /Einar
physical_devices = tf.config.experimental.list_physical_devices('GPU')
tf.config.experimental.set_memory_growth(physical_devices[0], True)
# Load data
(train_images,
train_labels), (test_images, test_labels) = datasets.get_dataset(DATASET_NAME)
# Preprocess
train_labels = tf.one_hot(train_labels.reshape((-1, )),
settings.DATASET_N_CLASSES[DATASET_NAME])
test_labels = tf.one_hot(test_labels.reshape((-1, )),
settings.DATASET_N_CLASSES[DATASET_NAME])
# Get model
model = cnn.get_model(dataset_name=DATASET_NAME, compile=True)
# Train
model.fit(train_images,
train_labels,
epochs=N_EPOCHS,
validation_data=(test_images, test_labels))
# Save weights
saveload.save_tf_model(model, "cnn")
def main():
from models import cnn
# u_net_cross_entropy
model = cnn.get_model(None, None, 3, do_compile=False)
model.compile(
optimizer='adam',
loss='binary_crossentropy',
metrics=['accuracy',
tf.keras.metrics.MeanIoU(num_classes=2), f1])
model_name = 'cnn_cross_entropy'
cross_val(model, model_name)
# dice
from losses import dice
loss = dice.dice_loss
model = cnn.get_model(None, None, 3, do_compile=False)
model.compile(
optimizer='adam',
loss=loss,
metrics=['accuracy',
tf.keras.metrics.MeanIoU(num_classes=2), f1])
model_name = 'cnn_dice'
cross_val(model, model_name)
# u_net_focal
from losses import focal
loss = focal.focal_loss
model = cnn.get_model(None, None, 3, do_compile=False)
model.compile(
optimizer='adam',
loss=loss,
metrics=['accuracy',
tf.keras.metrics.MeanIoU(num_classes=2), f1])
model_name = 'cnn_focal'
cross_val(model, model_name)
# u_net_lovasz
from losses import lovasz
loss = lovasz.lovasz_loss
model = cnn.get_model(None, None, 3, do_compile=False)
model.compile(
optimizer='adam',
loss=loss,
metrics=['accuracy',
tf.keras.metrics.MeanIoU(num_classes=2), f1])
model_name = 'cnn_lovasz'
cross_val(model, model_name)
# u_net_balanced_cross_entropy_class_weight
model = cnn.get_model(None, None, 3, do_compile=False)
model.compile(
optimizer='adam',
loss='binary_crossentropy',
metrics=['accuracy',
tf.keras.metrics.MeanIoU(num_classes=2), f1])
model_name = 'cnn_balanced_cross_entropy_class_weight'
cross_val(model, model_name, use_class_weight=True)
def main():
model = cnn.get_model(None, None, 3, do_compile=False)
model.load_weights('checkpoints/ckp_cnn_dice_SPECIALDATA.h5')
submission.create(model, 'ckp_cnn_dice_SPECIALDATA')