Classification Project (Colab)

1. 라이브러리 Import

import cv2
import numpy as np
import tensorflow as tf
from tensorflow import keras
import matplotlib.pyplot as plt
import imageio
from tensorflow.keras.preprocessing.image import ImageDataGenerator

2. Dataset 불러오기

url='https://github.com/******/******/blob/main/derm_dataset.zip?raw=true'
path_to_zip = keras.utils.get_file('dataset.zip', origin=url, extract=True, cache_dir='/content')


# 또는 아래와 같이 리눅스 명령어로 사용
# !mkdir "/content/dataset"
# !wget 'https://github.com/******/******/blob/main/dataset.zip?raw=true' -O "/content/dataset/dataset.zip"
# !unzip "/content/dataset/dataset.zip" -d "/content/dataset"

3. Train Set, Validation Set 지정(Diretory 통채로)

train_dir = "/content/datasets/train"
validation_dir = "/content/datasets/validation"

4.ImageDateGenerarator를 사용한 Augmentation

4-1. 객체 생성

• Image Tensor를 Min-Max Scaling (0~1값)
• Train Data Set을 Rotation, Shiftm, Zoom, Flip 등을 통한 Image Augmentation 수행
• Validation Set은 증식없이 진행

train_datagen = ImageDataGenerator(rescale=1./255, rotation_range=30, width_shift_range=0.1, height_shift_range=0.1, shear_range=0.1, zoom_range=0.1, horizontal_flip=True)
validation_datagen = ImageDataGenerator(rescale=1./255)

4-2. 객체 속성 지정

• Train set, Valdation set 연결
• 이미지 사이즈, 배치사이즈, Class Mode 지정

train_generator = train_datagen.flow_from_directory(
        train_dir,
        target_size=(128, 128),
        batch_size=32,
        shuffle=True,
        class_mode='categorical')

validation_generator = validation_datagen.flow_from_directory(
        validation_dir,
        target_size=(128, 128),
        batch_size=32,
        class_mode='categorical')

4-3. Generator로 부터 생성된 Image와 Label을 각각 x,y인자로 받음

x1, y1 = next(train_generator)
x2, y2 = next(validation_generator)

4-4. 결과 확인

print("train_generator 배치수:",len(train_generator))
print("validation_generator 배치수:",len(validation_generator))
print("train_generator (배치크기,x해상도,y해상도, Colorlayer):", x1.shape)
print("train_label (배치크기):", y1.shape)
print("validation_generator (배치크기,x해상도,y해상도, Colorlayer):", x2.shape)
print("validation_label (배치크기):", y2.shape)

stacked = np.hstack((x1[0], x1[1], x1[2], x1[3], x1[4]))

plt.figure(figsize=(20,4))
plt.imshow(stacked)

dict = {0:"acne", 1:"agespot",2:"flakyskin",3:"reddish"}
[dict[np.argmax(y1[i])] for i in range(0,5)]

5. 모델 설계

5-1. Backbone 모델 설정

• 모델 리스트 참고: https://www.tensorflow.org/api_docs/python/tf/keras/applications

#InceptionResNetV2 (Acc=0.8~0.9)
conv_base = tf.keras.applications.inception_resnet_v2.InceptionResNetV2(weights='imagenet', include_top=False, input_shape=(128, 128, 3))

#Exception (Acc=0.8~0.9)
#conv_base = tf.keras.applications.xception.Xception(weights='imagenet', include_top=False, input_shape=(128, 128, 3))

#VGG19 (Acc=0.6~0.7)
#conv_base = tf.keras.applications.vgg19.VGG19(weights='imagenet', include_top=False, input_shape=(128, 128, 3))

#MobileNetV2 (Acc=0.7~0.8)
#conv_base = tf.keras.applications.mobilenet_v2.MobileNetV2(weights='imagenet', include_top=False, input_shape=(128, 128, 3))

5-2. 분류기(Classifier) 설계

model = keras.Sequential()
model.add(conv_base)
model.add(keras.layers.Flatten())
model.add(keras.layers.Dense(256, activation='relu'))
model.add(keras.layers.Dense(4, activation='softmax'))

5-3. Backbone 모델의 학습된 Parameter는 추가 학습 진행 안하도록 설정

conv_base.trainable = False

5-4. 설계한 모델확인

model.summary()

5-5. Optimizer 설정

model.compile(loss='categorical_crossentropy',
              optimizer=tf.keras.optimizers.Nadam(learning_rate=0.001),
              metrics=['accuracy'])

6. 모델 학습

• 학습 Data set 지정: train_generator
• 배치수 지정(steps_per_epoch)
• 훈련횟수 지정(epochs)
• 검증데이터셋 및 배치수 지정

history = model.fit(
      train_generator,
      steps_per_epoch=len(train_generator),
      epochs=50,
      validation_data=validation_generator,
      validation_steps=len(validation_generator)
)

7. 학습결과 시각화

acc = history.history['accuracy']
val_acc = history.history['val_accuracy']
loss = history.history['loss']
val_loss = history.history['val_loss']

epochs = range(len(acc))

plt.plot(epochs, acc, 'bo', label='Training acc')
plt.plot(epochs, val_acc, 'b', label='Validation acc')
plt.title('Training and validation accuracy')
plt.legend()

plt.figure()

plt.plot(epochs, loss, 'bo', label='Training loss')
plt.plot(epochs, val_loss, 'b', label='Validation loss')
plt.title('Training and validation loss')
plt.legend()

plt.show()

8.  Prediction

# 테스트할 이미지 불러들여오기
url2 = "https://github.com/******/******/blob/main/predict.zip?raw=true"
path_to_zip = keras.utils.get_file('predict.zip', origin=url2, extract=True, cache_dir='/content',cache_subdir='datasets/test')
!rm /content/datasets/test/predict.zip

img_stack=[]
predict=[]
for i in range(1,5):
  test = f"/content/datasets/test/00{i}.jpg"
  img = imageio.imread(test)
  img = img/255
  res = model.predict(img.reshape(-1,128,128,3))
  img_stack.append(img)
  predict.append(dict[np.argmax(res)])
plt.figure(figsize=(20,4))
plt.imshow(np.hstack((img_stack[0:4])))
print(predict[0:4])