Question

训练vgg16时，训练和验证曲线会出现峰值，表示损失和准确性。我正在使用转移学习技术，并已更改了用于对性别进行分类的二元类问题的分类器。有人可以建议我为什么会出现这种尖峰，如何减少这种尖峰。

代码如下：

from keras.layers import Input, Lambda, Dense, Flatten, Dropout
from keras.models import Model
from keras.applications.vgg16 import VGG16
from keras.applications.vgg16 import preprocess_input
from keras.preprocessing import image
from keras.preprocessing.image import ImageDataGenerator
from keras.models import Sequential
import numpy as np
from glob import glob
import matplotlib.pyplot as plt

 # re-size all the images to this
 IMAGE_SIZE = [224, 224]

 train_path = 'E:/decompressed_images/data_set/train'
 valid_path = 'E:/decompressed_images/data_set/validation'

 # add preprocessing layer to the front of VGG
 vgg = VGG16(input_shape=IMAGE_SIZE + [3], weights='imagenet', include_top=False)

 # don't train existing weights
 for layer in vgg.layers:
   layer.trainable = False
# useful for getting number of classes
folders = glob('E:/decompressed_images/data_set/train*')
x = Flatten()(vgg.output)
# x = Dense(1000, activation='relu')(x)
prediction = Dense(len(folders), activation='sigmoid')(x)

# create a model object
model = Model(inputs=vgg.input, outputs=prediction)

# view the structure of the model
model.summary()

# tell the model what cost and optimization method to use
model.compile(
    loss='binary_crossentropy',
    optimizer='adam',
    metrics=['accuracy']
 )

from keras.preprocessing.image import ImageDataGenerator

 train_datagen = ImageDataGenerator(rescale = 1./255,
                              horizontal_flip = True,
                              vertical_flip = True,
                              width_shift_range = 0.1,
                              height_shift_range = 0.1,
                              zoom_range = 0.1,
                              rotation_range = 10)

   test_datagen = ImageDataGenerator(rescale = 1./255)

   training_set = train_datagen.flow_from_directory('E:/Ullu/new_trial__/balanced_dataset/train',
                                             target_size = (224, 224),
                                             batch_size = 64,
                                             class_mode = 'binary')

    test_set = test_datagen.flow_from_directory('E:/Ullu/new_trial__/balanced_dataset/test',
                                        target_size = (224, 224),
                                        batch_size = 64,
                                        class_mode = 'binary')

     r = model.fit_generator(
            training_set,
            validation_data=test_set,
            epochs=100,
            steps_per_epoch=len(training_set),
            validation_steps=len(test_set)
            )
      plt.plot(r.history['loss'], label='train loss')
      plt.plot(r.history['val_loss'], label='val loss')
      plt.legend()
      plt.show()
      plt.savefig('E:/Model_128_30/LossVal_loss.png')

     # accuracies
     plt.plot(r.history['accuracy'], label='train acc')
     plt.plot(r.history['val_accuracy'], label='val acc')
     plt.legend()
     plt.show()
     plt.savefig('E:/Model_128_30/AccVal_acc.png')

     import tensorflow as tf
     from keras.models import load_model
     model.save('E:/Model_128_30/128_30_wt.h5')

High and fluctuating training and validation accuracy image

High and fluctuating training and validation loss image

我尝试在最后一层使用dropout层（0.5），但我的准确性和训练与验证的损失是相同的。有人可以建议我哪里错了。谢谢。

训练vgg16时，训练和验证曲线的损失和准确性都有峰值

0 个答案: