Generate Image
이탤릭 볼드 이탤릭볼드
Workflow stages
- Question or problem definition.
- Acquire training and testing data.
- Wrangle, prepare, cleanse the data.
- Analyze, identify patterns, and explore the data.
- Model, predict and solve the problem.
- Visualize, report, and present the problem solving steps and final solution.
- Supply or submit the results.
기본적으로 설치되어 있어야하는 패키지는 아래 코드 를 사용한다.
import matplotlib.pyplot as plt
import numpy as np
from sklearn.model_selection import train_test_split
from skimage.io import imread
from skimage.transform import pyramid_reduce, resize
import os, glob
from keras.layers import Input, Activation, Conv2D, Flatten, Dense, MaxPooling2D, Dropout, Add, LeakyReLU, UpSampling2D
from keras.models import Model, load_model
from keras.callbacks import ReduceLROnPlateau
Data 전처리
img_list = sorted(glob.glob('2d_images/*.tif'))
mask_list = sorted(glob.glob('2d_masks/*.tif'))
IMG_SIZE = 256
x_data, y_data = np.empty((2, len(img_list), IMG_SIZE, IMG_SIZE, 1), dtype=np.float32)
for i, img_path in enumerate(img_list):
img = imread(img_path)
img = resize(img, output_shape=(IMG_SIZE, IMG_SIZE, 1), preserve_range=True)
x_data[i] = img
for i, img_path in enumerate(mask_list):
img = imread(img_path)
img = resize(img, output_shape=(IMG_SIZE, IMG_SIZE, 1), preserve_range=True)
y_data[i] = img
y_data /= 255.
fig, ax = plt.subplots(1, 2)
ax[0].imshow(x_data[12].squeeze(), cmap='gray')
ax[1].imshow(y_data[12].squeeze(), cmap='gray')
x_train, x_val, y_train, y_val = train_test_split(x_data, y_data, test_size=0.1)
np.save('dataset/x_train.npy', x_train)
np.save('dataset/y_train.npy', y_train)
np.save('dataset/x_val.npy', x_val)
np.save('dataset/y_val.npy', y_val)
Load Dataset
Encoder 차원 축소를 통해 핵심요소 추출 (DownSampling, ex MaxPooling2D)
Decoder 압축된 정보로부터 차원 확장을 통해 원하는 정보로 복원 ( UpsSampling , Upsampling2D)
x_train = np.load('dataset/x_train.npy') # CT
y_train = np.load('dataset/y_train.npy') # Mask
x_val = np.load('dataset/x_val.npy')s # CT정답
y_val = np.load('dataset/y_val.npy') # Mask 정답
모델 사용
dinputs = Input(shape=(256, 256, 1))
net = Conv2D(32, kernel_size=3, activation='relu', padding='same')(inputs)
net = MaxPooling2D(pool_size=2, padding='same')(net)
net = Conv2D(64, kernel_size=3, activation='relu', padding='same')(net)
net = MaxPooling2D(pool_size=2, padding='same')(net)
net = Conv2D(128, kernel_size=3, activation='relu', padding='same')(net)
net = MaxPooling2D(pool_size=2, padding='same')(net)
net = Dense(128, activation='relu')(net)
net = UpSampling2D(size=2)(net)
net = Conv2D(128, kernel_size=3, activation='sigmoid', padding='same')(net)
net = UpSampling2D(size=2)(net)
net = Conv2D(64, kernel_size=3, activation='sigmoid', padding='same')(net)
net = UpSampling2D(size=2)(net)
outputs = Conv2D(1, kernel_size=3, activation='sigmoid', padding='same')(net)
model = Model(inputs=inputs, outputs=outputs)
model.compile(optimizer='adam', loss='binary_crossentropy', metrics=['acc', 'mse'])
model.summary()
# tf 2.x 부터는
model = models.Sequential()
model.add(layers.Conv2D(32, kernel_size=3 , activation='relu', padding='same', input_shape=(256, 256, 1)))
model.add(layers.MaxPooling2D())
model.add(layers.Conv2D(64, kernel_size=3, activation='relu', padding='same'))
model.add(layers.MaxPooling2D())
model.add(layers.Conv2D(128, kernel_size=3, activation='relu', padding='same'))
model.add(layers.MaxPooling2D())
model.add(layers.Dense(128, activation='relu'))
model.add(layers.UpSampling2D(size=2))
model.add(layers.Conv2D(128, kernel_size=3, activation='sigmoid', padding='same'))
model.add(layers.UpSampling2D(size=2))
model.add(layers.Conv2D(64, kernel_size=3, activation='sigmoid', padding='same'))
model.add(layers.UpSampling2D(size=2))
model.add(layers.Conv2D(1, kernel_size=3 , activation='sigmoid', padding='same'))
model.summary()
Train
history = model.fit(x_train, y_train, validation_data=(x_val, y_val), epochs=100, batch_size=32, callbacks=[
ReduceLROnPlateau(monitor='val_loss', factor=0.2, patience=10, verbose=1, mode='auto', min_lr=1e-05)
])
Evaluation
fig, ax = plt.subplots(2, 2, figsize=(10, 7))
ax[0, 0].set_title('loss')
ax[0, 0].plot(history.history['loss'], 'r')
ax[0, 1].set_title('acc')
ax[0, 1].plot(history.history['acc'], 'b')
ax[1, 0].set_title('val_loss')
ax[1, 0].plot(history.history['val_loss'], 'r--')
ax[1, 1].set_title('val_acc')
ax[1, 1].plot(history.history['val_acc'], 'b--')
