import os
import tensorflow as tf
from tensorflow.keras import Sequential, regularizers
from tensorflow.keras.layers import MaxPooling2D, Conv2D, Flatten, Dense, Dropout, Activation, \
BatchNormalization
from tensorflow.keras.optimizers import Adam, SGD
from tensorflow.keras.preprocessing.image import ImageDataGenerator
from tensorflow.keras.applications.vgg16 import VGG16
from tensorflow_core.python.keras import Model
if __name__ == '__main__':
os.environ["KMP_DUPLICATE_LIB_OK"] = "TRUE"
image_generator = ImageDataGenerator(rescale=1. / 255)
image_generator2 = ImageDataGenerator(rescale=1. / 255)
train_data = image_generator.flow_from_directory("/Users/kongbai/temp/data/train", batch_size=20,
color_mode='grayscale',
shuffle=False,
#
save_to_dir="/Users/kongbai/temp/data/gen/t",
save_prefix='train_',
target_size=(100, 100), class_mode="categorical")
test_data = image_generator.flow_from_directory("/Users/kongbai/temp/data/test", batch_size=1,
color_mode='grayscale',
shuffle=False,
# save_to_dir=SAVE_PATH,
#
# save_prefix='gen',
target_size=(100, 100), class_mode='categorical')
vailted_data = image_generator.flow_from_directory("/Users/kongbai/temp/data/vailted", batch_size=20,
color_mode='grayscale',
shuffle=True,
save_to_dir="/Users/kongbai/temp/data/gen/v",
save_prefix='vailted_',
target_size=(100, 100), class_mode='categorical')
model = Sequential()
model.add(Conv2D(32, 3, 1, input_shape=(100, 100, 1), activation='relu'))
model.add(MaxPooling2D(3))
model.add(Conv2D(128, 1, padding='same'))
model.add(Conv2D(64, 1))
model.add(Dropout(0.1))
model.add(Flatten())
model.add(Dense(512, activation='relu'))
model.add(Dropout(0.4))
model.add(Dense(2, activation='softmax', ))
model.compile(loss=tf.keras.losses.CategoricalCrossentropy(from_logits=False),
optimizer=tf.keras.optimizers.SGD(1e-4),
metrics=['accuracy'])
model.summary()
model.fit_generator(train_data, shuffle=True, steps_per_epoch=20, epochs=20, workers=16,
validation_data=vailted_data,
validation_steps=20)
model.save("demo.h5")
训练的acc一直在0.5,loss直接固定在0.69了...............这就是个识别一张照片上的两个人是否是同一个人的网络呀,也没什么东西了呀 损失函数也改过Adam 也是一样的 有大佬帮忙看看吗?
帮你改了下代码，路径你自己再改改，验证集acc能跑到0.8以上了。
import os
import tensorflow as tf
from tensorflow.keras import Sequential, regularizers
from tensorflow.keras.layers import MaxPooling2D, Conv2D, Flatten, Dense, Dropout, Activation, \
BatchNormalization
from tensorflow.keras.optimizers import Adam, SGD
from tensorflow.keras.preprocessing.image import ImageDataGenerator
from tensorflow.keras.applications.resnet50 import ResNet50
if __name__ == '__main__':
os.environ["KMP_DUPLICATE_LIB_OK"] = "TRUE"
image_generator = ImageDataGenerator(rescale=1. / 255)
image_generator2 = ImageDataGenerator(rescale=1. / 255)
train_data = image_generator.flow_from_directory("mg", batch_size=32,
color_mode='grayscale',
shuffle=True,
# #
# save_to_dir="/Users/kongbai/temp/data/gen/t",
#
# save_prefix='train_',
target_size=(256, 256), class_mode="categorical")
test_data = image_generator.flow_from_directory("mg_test", batch_size=32,
color_mode='grayscale',
shuffle=True,
# save_to_dir=SAVE_PATH,
#
# save_prefix='gen',
target_size=(256, 256), class_mode='categorical')
vailted_data = image_generator.flow_from_directory("mg_test", batch_size=32,
color_mode='grayscale',
shuffle=True,
target_size=(256, 256), class_mode='categorical')
model = Sequential()
model.add(Conv2D(filters=32, kernel_size=(5, 5), strides=(2, 2), \
input_shape=(256, 256, 1)))
model.add(Activation('relu'))
model.add(MaxPooling2D(pool_size=(2, 2)))
model.add(Conv2D(filters=64, kernel_size=(3, 3), strides=(1, 1)))
model.add(Activation('relu'))
model.add(MaxPooling2D(pool_size=(2, 2)))
# model.add(Conv2D(32, (3, 3)))
# model.add(Activation('relu'))
# model.add(MaxPooling2D(pool_size=(2, 2)))
#
# model.add(Conv2D(64, (3, 3)))
# model.add(Activation('relu'))
# model.add(MaxPooling2D(pool_size=(2, 2)))
model.add(Flatten())
model.add(Dropout(0.8))
model.add(Dense(128))
model.add(Activation('relu'))
model.add(Dropout(0.8)) # Dropout防止过拟合
model.add(Dense(2))
model.add(Activation('sigmoid'))
# model = tf.keras.applications.ResNet50(input_shape=(224,224, 1), classes=2, weights=None)
model.compile(optimizer=tf.keras.optimizers.Adam(), loss="categorical_crossentropy", metrics=["acc"], )
# model.compile(loss=tf.keras.losses.CategoricalCrossentropy(from_logits=False),
# optimizer=tf.keras.optimizers.SGD(1e-4),
# metrics=['accuracy'])
model.summary()
model.fit_generator(train_data, shuffle=True, epochs=10, workers=16,
validation_data=vailted_data)
model.save("demo.h5")
根据经验，如果训练数据不是很多的话，那就过拟合了，建议把模型结构设计的再简单点。