from plot import plot_learning_curve
from metrics import dice_coef_loss, dice_coef
# Read the data
path = '/Lab1/Lab3/X_ray/'
img_h, img_w = 256, 256
Mask = gen_list(path, 'Mask')
Img = gen_list(path,'Image')
Mask_train, Mask_val, Img_train, Img_val = shuffle_split(Mask, Img, 0.8)
Mask_train = read_data(path+'Mask/', Mask_train, img_h, img_w)
Mask_val = read_data(path+'Mask/', Mask_val, img_h, img_w)
Img_train = read_data(path+'Image/', Img_train, img_h, img_w)
Img_val = read_data(path+'Image/', Img_val, img_h, img_w)
# Train the model
model = get_UNet(img_shape=(256,256,1), Base=16, depth=4, inc_rate=2,
activation='relu', drop=0.5, batchnorm=True)
model.compile(optimizer=Adam(lr=0.0001), loss='binary_crossentropy',
metrics=[dice_coef])
History = model.fit(Img_train, Mask_train, batch_size=8, epochs=150, verbose=2,
validation_data=(Img_val, Mask_val))
# Plot the learning curve
plot_learning_curve(History, 'Task1a')
# Train the model
model = get_UNet(img_shape=(256,256,1), Base=16, depth=4, inc_rate=2,
activation='relu', drop=0.5, batchnorm=True)
model.compile(optimizer=Adam(lr=0.0001), loss=[dice_coef_loss],
# Get the patches of training images
patches_imgs_train, patches_masks_train = prepare_training_data(
imgs_train, truth_train, patch_h, patch_w, N_patches)
# Parameters for training model
N_epochs = 150
batch_size = 64
lr = 0.1
decay_rate = lr / N_epochs
sgd = SGD(lr=lr, momentum=0.8, decay=decay_rate, nesterov=False)
model = get_UNet(img_shape=(patch_h, patch_w, 1),
Base=32,
depth=4,
inc_rate=2,
activation='relu',
drop=0.2,
batchnorm=True,
N=2)
model.compile(optimizer='sgd',
loss='categorical_crossentropy',
metrics=['accuracy'])
# Save the architecture of the model
json_string = model.to_json()
open('model_architecture.json', 'w').write(json_string)
# Save the best weights
checkpointer = ModelCheckpoint(filepath='best_weights.h5',
monitor='val_loss',
k=3 # k=5
Mask = split_list(Mask,k)
Img = split_list(Img,k)
for i in range(k):
Mask_val = list(Mask[i])
Mask_train =list(chain.from_iterable(Mask[:i] + Mask[i+1:]))
Img_val = list(Img[i])
Img_train =list(chain.from_iterable(Img[:i] + Img[i+1:]))
Mask_train = read_data(path+'Mask/', Mask_train, img_h, img_w)
Mask_val =read_data(path+'Mask/', Mask_val, img_h, img_w)
Img_train = read_data(path+'Image/', Img_train, img_h, img_w)
Img_val = read_data(path+'Image/', Img_val, img_h, img_w)
model = get_UNet(img_shape=(img_h, img_w, 1), Base=16, depth=4, inc_rate=2,
activation='relu', drop=0, batchnorm=True, N=2, weight_use=False)
model.compile(optimizer=Adam(lr=1e-5), loss=[dice_coef_loss],
metrics=[dice_coef,precision,recall])
History = model.fit(Img_train, Mask_train, batch_size=8, epochs=150, verbose=1,
validation_data=(Img_val, Mask_val))
plot_learning_curve(History, 'Task1_k={0}_loss_{1}_'.format(k,i+1))
plot_validation_metric(History, 'Task1_k={0}_metrics_{1}_'.format(k,i+1))
''' Task 2 '''
k=3 # k=5
Mask = split_list(Mask,k)
data_gen_args = dict(rotation_range=5,
width_shift_range=0.1,
height_shift_range=0.1,
validation_split=0.2)
image_train_datagen = ImageDataGenerator(**data_gen_args)
mask_train_datagen = ImageDataGenerator(**data_gen_args)
image_val_datagen = ImageDataGenerator(**data_gen_args)
mask_val_datagen = ImageDataGenerator(**data_gen_args)
seed = 1
batch_size = 8
model = get_UNet(img_shape=(img_h, img_w, 1),
Base=16,
depth=4,
inc_rate=2,
activation='relu',
drop=0,
batchnorm=True,
N=2)
model.compile(optimizer=Adam(lr=0.0001),
loss=[dice_coef_loss],
metrics=[dice_coef, recall, precision])
# Train model
History = model.fit_generator(
XYaugmentGenerator(image_train_datagen, mask_train_datagen, Img_train,
Mask_train, seed, batch_size),
steps_per_epoch=np.ceil(float(len(Img_train)) / float(batch_size)),
validation_data=XYaugmentGenerator(image_val_datagen, mask_val_datagen,
Img_val, Mask_val, seed, batch_size),