def test_generator_1():
# ### CONSTANTS
HEIGHT, WIDTH, CHANNELS, IMG_COLOR_MODE, MSK_COLOR_MODE, NUM_CLASS, \
KS1, KS2, KS3, DL1, DL2, DL3, NF, NFL, NR1, NR2, DIL_MODE, W_MODE, LS, \
TRAIN_SIZE, VAL_SIZE, TEST_SIZE, DR1, DR2, CLASSES, IMG_CLASS = _Params()
TRAIN_IMG_PATH, TRAIN_MSK_PATH, TRAIN_MSK_CLASS, VAL_IMG_PATH, \
VAL_MSK_PATH, VAL_MSK_CLASS, TEST_IMG_PATH, TEST_MSK_PATH, TEST_MSK_CLASS = _Paths()
TRAIN_SEED, VAL_SEED, TEST_SEED = _Seeds()
test_data_gen_img_args = dict(rescale=1. / 255)
if MSK_COLOR_MODE == 'rgb':
test_data_gen_msk_args = dict(rescale=1. / 255)
elif MSK_COLOR_MODE == 'grayscale':
test_data_gen_msk_args = dict(preprocessing_function=to_test_indices)
test_image_datagen = ImageDataGenerator(**test_data_gen_img_args)
test_mask_datagen = ImageDataGenerator(**test_data_gen_msk_args)
test_image_generator = test_image_datagen.flow_from_directory(
TEST_IMG_PATH,
target_size=(HEIGHT, WIDTH),
classes=[IMG_CLASS],
color_mode=IMG_COLOR_MODE,
class_mode=None,
batch_size=TEST_SIZE,
shuffle=False,
seed=TEST_SEED)
test_mask_generator = test_mask_datagen.flow_from_directory(
TEST_MSK_PATH,
target_size=(HEIGHT, WIDTH),
classes=CLASSES,
color_mode=MSK_COLOR_MODE,
class_mode=None,
batch_size=TEST_SIZE,
shuffle=False,
seed=TEST_SEED)
while True:
if MSK_COLOR_MODE == 'rgb':
yield (test_image_generator.next(), test_mask_generator.next())
elif MSK_COLOR_MODE == 'grayscale':
yield(test_image_generator.next(), to_one_hot_val(test_mask_generator.next(), \
TRAIN_SIZE,HEIGHT,WIDTH,NUM_CLASS))
def test_generator_2():
# ### CONSTANTS
HEIGHT, WIDTH, CHANNELS, IMG_COLOR_MODE, MSK_COLOR_MODE, NUM_CLASS, \
KS1, KS2, KS3, DL1, DL2, DL3, NF, NFL, NR1, NR2, DIL_MODE, W_MODE, LS, \
TRAIN_SIZE, VAL_SIZE, TEST_SIZE, DR1, DR2, CLASSES, IMG_CLASS = _Params()
TRAIN_IMG_PATH, TRAIN_MSK_PATH, TRAIN_MSK_CLASS, VAL_IMG_PATH, \
VAL_MSK_PATH, VAL_MSK_CLASS, TEST_IMG_PATH, TEST_MSK_PATH, TEST_MSK_CLASS = _Paths()
TRAIN_SEED, VAL_SEED, TEST_SEED = _Seeds()
test_data_gen_img_args = dict(rescale=1. / 255)
test_data_gen_msk_args = dict(rescale=1. / 255)
test_image_datagen = ImageDataGenerator(**test_data_gen_img_args)
test_mask_datagen = ImageDataGenerator(**test_data_gen_msk_args)
test_image_generator = test_image_datagen.flow_from_directory(
TEST_IMG_PATH,
target_size=(HEIGHT, WIDTH),
classes=[IMG_CLASS],
color_mode=IMG_COLOR_MODE,
class_mode=None,
batch_size=TEST_SIZE,
shuffle=False,
seed=TEST_SEED,
# save_to_dir='dataset/val_local/images/save',
)
for i in range(NUM_CLASS):
globals()['test_mask_gen_{}'.format(
i)] = test_mask_datagen.flow_from_directory(
TEST_MSK_PATH,
target_size=(HEIGHT, WIDTH),
classes=[CLASSES[i]],
color_mode=MSK_COLOR_MODE,
class_mode=None,
batch_size=TEST_SIZE,
shuffle=False,
seed=TEST_SEED,
# save_to_dir='dataset/val_local/masks/save',
)
while True:
yield(test_image_generator.next(), \
np.round(np.squeeze(np.stack([globals()['test_mask_gen_{}'.format(i)].next() for i in range(NUM_CLASS)],axis=3))))
from MODULES.Losses import weighted_tani_loss
import tensorflow as tf
from tensorflow.keras.models import Model
from tensorflow.keras.layers import Input, Activation, SeparableConv2D
from tensorflow.keras.layers import Conv2D, Concatenate, Add, LeakyReLU
from tensorflow.keras.layers import Dropout, SpatialDropout2D
from tensorflow.keras.initializers import he_normal
from tensorflow.keras.optimizers import Adam
# In[2]
# ### CONSTANTS
HEIGHT, WIDTH, CHANNELS, IMG_COLOR_MODE, MSK_COLOR_MODE, NUM_CLASS, \
KS1, KS2, KS3, DL1, DL2, DL3, NF, NFL, NR1, NR2, DIL_MODE, W_MODE, LS, \
TRAIN_SIZE, VAL_SIZE, TEST_SIZE, DR1, DR2, CLASSES, IMG_CLASS = _Params()
TRAIN_IMG_PATH, TRAIN_MSK_PATH, TRAIN_MSK_CLASS, VAL_IMG_PATH, \
VAL_MSK_PATH, VAL_MSK_CLASS, TEST_IMG_PATH, TEST_MSK_PATH, TEST_MSK_CLASS = _Paths()
# In[3]:
# ### RESNET-ATROUS
def ResNet_Atrous(input_shape=(HEIGHT, WIDTH, CHANNELS),
num_class=NUM_CLASS,
ks1=KS1, ks2=KS2, ks3=KS3,
dl1=DL1, dl2=DL2, dl3=DL3,
filters=NF,resblock1=NR1,
r_filters=NFL, resblock2=NR2,
dil_mode=DIL_MODE,
def train_generator_1():
# ### CONSTANTS
HEIGHT, WIDTH, CHANNELS, IMG_COLOR_MODE, MSK_COLOR_MODE, NUM_CLASS, \
KS1, KS2, KS3, DL1, DL2, DL3, NF, NFL, NR1, NR2, DIL_MODE, W_MODE, LS, \
TRAIN_SIZE, VAL_SIZE, TEST_SIZE, DR1, DR2, CLASSES, IMG_CLASS = _Params()
TRAIN_IMG_PATH, TRAIN_MSK_PATH, TRAIN_MSK_CLASS, VAL_IMG_PATH, \
VAL_MSK_PATH, VAL_MSK_CLASS, TEST_IMG_PATH, TEST_MSK_PATH, TEST_MSK_CLASS = _Paths()
TRAIN_SEED, VAL_SEED, TEST_SEED = _Seeds()
train_data_gen_img_args = dict(rescale=1. / 255,
horizontal_flip=True,
vertical_flip=True,
rotation_range=90,
width_shift_range=0.1,
height_shift_range=0.1,
shear_range=15,
zoom_range=0.1,
fill_mode='reflect')
if MSK_COLOR_MODE == 'rgb':
train_data_gen_msk_args = dict(rescale=1. / 255,
horizontal_flip=True,
vertical_flip=True,
rotation_range=90,
width_shift_range=0.1,
height_shift_range=0.1,
shear_range=15,
zoom_range=0.1,
fill_mode='reflect')
elif MSK_COLOR_MODE == 'grayscale':
train_data_gen_msk_args = dict(preprocessing_function=to_train_indices,
horizontal_flip=True,
vertical_flip=True,
rotation_range=90,
width_shift_range=0.1,
height_shift_range=0.1,
shear_range=15,
zoom_range=0.1,
fill_mode='reflect')
train_image_datagen = ImageDataGenerator(**train_data_gen_img_args)
train_mask_datagen = ImageDataGenerator(**train_data_gen_msk_args)
train_image_generator = train_image_datagen.flow_from_directory(
TRAIN_IMG_PATH,
target_size=(HEIGHT, WIDTH),
classes=[IMG_CLASS],
color_mode=IMG_COLOR_MODE,
class_mode=None,
batch_size=TRAIN_SIZE,
shuffle=False,
seed=TRAIN_SEED)
train_mask_generator = train_mask_datagen.flow_from_directory(
TRAIN_MSK_PATH,
target_size=(HEIGHT, WIDTH),
classes=CLASSES,
color_mode=MSK_COLOR_MODE,
class_mode=None,
batch_size=TRAIN_SIZE,
shuffle=False,
seed=TRAIN_SEED)
while True:
if MSK_COLOR_MODE == 'rgb':
yield (train_image_generator.next(), train_mask_generator.next())
elif MSK_COLOR_MODE == 'grayscale':
yield(train_image_generator.next(), to_one_hot_train(train_mask_generator.next(), \
TRAIN_SIZE,HEIGHT,WIDTH,NUM_CLASS))
def train_generator_2():
global mask
# ### CONSTANTS
HEIGHT, WIDTH, CHANNELS, IMG_COLOR_MODE, MSK_COLOR_MODE, NUM_CLASS, \
KS1, KS2, KS3, DL1, DL2, DL3, NF, NFL, NR1, NR2, DIL_MODE, W_MODE, LS, \
TRAIN_SIZE, VAL_SIZE, TEST_SIZE, DR1, DR2, CLASSES, IMG_CLASS = _Params()
TRAIN_IMG_PATH, TRAIN_MSK_PATH, TRAIN_MSK_CLASS, VAL_IMG_PATH, \
VAL_MSK_PATH, VAL_MSK_CLASS, TEST_IMG_PATH, TEST_MSK_PATH, TEST_MSK_CLASS = _Paths()
TRAIN_SEED, VAL_SEED, TEST_SEED = _Seeds()
train_data_gen_img_args = dict(rescale=1. / 255,
horizontal_flip=True,
vertical_flip=True,
rotation_range=90,
width_shift_range=0.1,
height_shift_range=0.1,
shear_range=15,
zoom_range=0.1,
fill_mode='reflect')
train_data_gen_msk_args = dict(rescale=1. / 255,
horizontal_flip=True,
vertical_flip=True,
rotation_range=90,
width_shift_range=0.1,
height_shift_range=0.1,
shear_range=15,
zoom_range=0.1,
fill_mode='reflect')
train_image_datagen = ImageDataGenerator(**train_data_gen_img_args)
train_mask_datagen = ImageDataGenerator(**train_data_gen_msk_args)
train_image_generator = train_image_datagen.flow_from_directory(
TRAIN_IMG_PATH,
target_size=(HEIGHT, WIDTH),
classes=[IMG_CLASS],
color_mode=IMG_COLOR_MODE,
class_mode=None,
batch_size=TRAIN_SIZE,
shuffle=False,
seed=TRAIN_SEED,
# save_to_dir='dataset/train_local/images/save',
)
for i in range(NUM_CLASS):
globals()['train_mask_gen_{}'.format(
i)] = train_mask_datagen.flow_from_directory(
TRAIN_MSK_PATH,
target_size=(HEIGHT, WIDTH),
classes=[CLASSES[i]],
color_mode=MSK_COLOR_MODE,
class_mode=None,
batch_size=TRAIN_SIZE,
shuffle=False,
seed=TRAIN_SEED,
# save_to_dir='dataset/train_local/masks/save',
)
while True:
yield(train_image_generator.next(), \
np.round(np.squeeze(np.stack([globals()['train_mask_gen_{}'.format(i)].next() for i in range(NUM_CLASS)],axis=3))))