예제
import tensorflow as tf
from tensorflow.keras.models import Sequential
from tensorflow.keras.layers import Dense, Conv2D, MaxPoolling2D
import numpy as pd
img_width = 225
img_height = 225
batch_size = 10
epochs = 50
num_classes = 2
# 파일에 저장된 이미지를 학습용 데이터로 로드한다
train_ds = tf.keras.preprocessing.image_dataset_from_directory(
'images',seed=123,image_size=(img_height,img_width),batch_size=batch_size
)
# Found 20 files belonging to 2 classes.
class_names = train_ds.class_names # ['bottle', 'cup']
model = Sequential([
tf.keras.layers.experimental.preprocessing.Rescaling(1.0/255),
Conv2D(32,3,activation='relu'), # 총 필터의 갯수 32, 필터의 사이즈 3x3, 활성화 함수 relu 음수는 0 양수는 그대로
MaxPooling2D(),
Conv2D(32,3,activation='relu'),
MaxPooling2D(),
Conv2D(32,3,activation='relu'),
MaxPooling2D(),
Flatten(),
Dense(128, activation='relu'),
Dense(num_classes, activation='softmax')
])
model.compile(loss='sparse_categorical_crossentropy',
optimizer='adam',
metrics=['accuracy'])
hist = model.fit(train_ds,epochs=epochs)
loss = hist.history['loss']
accuracy = hist.history['accuracy']
# 시각화
import matplotlib.pyplot as plt
plt.figure(figsize=(5,2))
plt.plot(loss, 'r--', label='loss')
plt.plot(accuracy, 'g', label='accuracy')
plt.xlabel('Epoch')
plt.ylabel('Error')
plt.legend()
plt.show()
.png)
# 테스트 이미지 준비
test_image_path = 'images_test/cup_test/r1ti21VgQf-5slrBHv4fAA.jpg'
img = tf.keras.preprocessing.image.load_img(test_image_path, target_size=(img_height,img_width))
# 신경망에 전달할 수 있는 데이터로 변환
img_array = tf.keras.preprocessing.image.img_to_array(img)
img_array.shape # (225, 225, 3)
img_array2 = tf.expand_dims(img_array, 0) # data, axis
pred = model.predict(img_array2) # [0.000014, 0.999986]
percent = np.max(pred)*100 # 99.99856948852539
idx = np.argmax(pred) # 1, 제일 큰 인덱스를 리턴
class_names[idx], f'{percent}%' # '('cup', '99.99856948852539%')